Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?
Identifieur interne : 004158 ( Istex/Corpus ); précédent : 004157; suivant : 004159Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?
Auteurs : Pawan Sinha ; Amy A. KaliaSource :
- Seeing and Perceiving [ 1878-4755 ] ; 2012.
Url:
DOI: 10.1163/187847511X584443
Links to Exploration step
ISTEX:116341E5C4AB855FACAAB4DBF5D84AA2479A0825Le document en format XML
<record><TEI wicri:istexFullTextTei="biblStruct"><teiHeader><fileDesc><titleStmt><title>Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title><author><name sortKey="Sinha, Pawan" sort="Sinha, Pawan" uniqKey="Sinha P" first="Pawan" last="Sinha">Pawan Sinha</name></author><author><name sortKey="Kalia, Amy A" sort="Kalia, Amy A" uniqKey="Kalia A" first="Amy A." last="Kalia">Amy A. Kalia</name></author></titleStmt><publicationStmt><idno type="wicri:source">ISTEX</idno><idno type="RBID">ISTEX:116341E5C4AB855FACAAB4DBF5D84AA2479A0825</idno><date when="2012" year="2012">2012</date><idno type="doi">10.1163/187847511X584443</idno><idno type="url">https://api.istex.fr/document/116341E5C4AB855FACAAB4DBF5D84AA2479A0825/fulltext/pdf</idno><idno type="wicri:Area/Istex/Corpus">004158</idno></publicationStmt><sourceDesc><biblStruct><analytic><title level="a">Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title><author><name sortKey="Sinha, Pawan" sort="Sinha, Pawan" uniqKey="Sinha P" first="Pawan" last="Sinha">Pawan Sinha</name></author><author><name sortKey="Kalia, Amy A" sort="Kalia, Amy A" uniqKey="Kalia A" first="Amy A." last="Kalia">Amy A. Kalia</name></author></analytic><monogr></monogr><series><title level="j">Seeing and Perceiving</title><title level="j" type="sub">A Journal of Multisensory Science</title><title level="j" type="abbrev">SP</title><idno type="ISSN">1878-4755</idno><idno type="eISSN">1878-4763</idno><imprint><publisher>BRILL</publisher><pubPlace>The Netherlands</pubPlace><date type="published" when="2012">2012</date><biblScope unit="volume">25</biblScope><biblScope unit="issue">3-4</biblScope><biblScope unit="page" from="287">287</biblScope><biblScope unit="page" to="302">302</biblScope></imprint><idno type="ISSN">1878-4755</idno></series><idno type="istex">116341E5C4AB855FACAAB4DBF5D84AA2479A0825</idno><idno type="DOI">10.1163/187847511X584443</idno><idno type="href">18784763_025_03-04_s004_text.pdf</idno></biblStruct></sourceDesc><seriesStmt><idno type="ISSN">1878-4755</idno></seriesStmt></fileDesc><profileDesc><textClass></textClass><langUsage><language ident="en">en</language></langUsage></profileDesc></teiHeader></TEI><istex><corpusName>brill-journals</corpusName><author><json:item><name>Pawan Sinha</name></json:item><json:item><name>Amy A. Kalia</name></json:item></author><subject><json:item><lang><json:string>eng</json:string></lang><value>haptic</value></json:item><json:item><lang><json:string>eng</json:string></lang><value>shape acquisition</value></json:item><json:item><lang><json:string>eng</json:string></lang><value>Tactile picture perception</value></json:item></subject><language><json:string>eng</json:string></language><qualityIndicators><score>5.512</score><pdfVersion>1.4</pdfVersion><pdfPageSize>453.543 x 680.315 pts</pdfPageSize><refBibsNative>false</refBibsNative><keywordCount>3</keywordCount><abstractCharCount>0</abstractCharCount><pdfWordCount>5411</pdfWordCount><pdfCharCount>31984</pdfCharCount><pdfPageCount>16</pdfPageCount><abstractWordCount>1</abstractWordCount></qualityIndicators><title>Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title><genre.original><json:string>research-article</json:string></genre.original><genre><json:string>research-article</json:string></genre><host><volume>25</volume><pages><last>302</last><first>287</first></pages><issn><json:string>1878-4755</json:string></issn><issue>3-4</issue><genre><json:string>Journal</json:string></genre><language><json:string>unknown</json:string></language><eissn><json:string>1878-4763</json:string></eissn><title>Seeing and Perceiving</title></host><publicationDate>2012</publicationDate><doi><json:string>10.1163/187847511X584443</json:string></doi><id>116341E5C4AB855FACAAB4DBF5D84AA2479A0825</id><score>1</score><fulltext><json:item><original>true</original><mimetype>application/pdf</mimetype><extension>pdf</extension><uri>https://api.istex.fr/document/116341E5C4AB855FACAAB4DBF5D84AA2479A0825/fulltext/pdf</uri></json:item><json:item><original>false</original><mimetype>application/zip</mimetype><extension>zip</extension><uri>https://api.istex.fr/document/116341E5C4AB855FACAAB4DBF5D84AA2479A0825/fulltext/zip</uri></json:item><json:item><original>false</original><mimetype>text/plain</mimetype><extension>txt</extension><uri>https://api.istex.fr/document/116341E5C4AB855FACAAB4DBF5D84AA2479A0825/fulltext/txt</uri></json:item><istex:fulltextTEI uri="https://api.istex.fr/document/116341E5C4AB855FACAAB4DBF5D84AA2479A0825/fulltext/tei"><teiHeader><fileDesc><titleStmt><title level="a">Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title><respStmt xml:id="ISTEX-API" resp="Références bibliographiques récupérées via GROBID" name="ISTEX-API (INIST-CNRS)"></respStmt></titleStmt><publicationStmt><authority>ISTEX</authority><publisher>BRILL</publisher><pubPlace>The Netherlands</pubPlace><availability><p>BRILL Journals</p></availability><date>2012</date></publicationStmt><sourceDesc><biblStruct type="inbook"><analytic><title level="a">Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title><author><persName><forename type="first">Pawan</forename><surname>Sinha</surname></persName></author><author><persName><forename type="first">Amy A.</forename><surname>Kalia</surname></persName></author></analytic><monogr><title level="j">Seeing and Perceiving</title><title level="j" type="sub">A Journal of Multisensory Science</title><title level="j" type="abbrev">SP</title><idno type="pISSN">1878-4755</idno><idno type="eISSN">1878-4763</idno><imprint><publisher>BRILL</publisher><pubPlace>The Netherlands</pubPlace><date type="published" when="2012"></date><biblScope unit="volume">25</biblScope><biblScope unit="issue">3-4</biblScope><biblScope unit="page" from="287">287</biblScope><biblScope unit="page" to="302">302</biblScope></imprint></monogr><idno type="istex">116341E5C4AB855FACAAB4DBF5D84AA2479A0825</idno><idno type="DOI">10.1163/187847511X584443</idno><idno type="href">18784763_025_03-04_s004_text.pdf</idno></biblStruct></sourceDesc></fileDesc><profileDesc><creation><date>2012</date></creation><langUsage><language ident="en">en</language></langUsage><textClass><keywords scheme="keyword"><list><head>Keywords</head><item><term>haptic</term></item><item><term>shape acquisition</term></item><item><term>Tactile picture perception</term></item></list></keywords></textClass></profileDesc><revisionDesc><change when="2012">Created</change><change when="2012">Published</change><change xml:id="refBibs-istex" who="#ISTEX-API" when="2016-3-13">References added</change></revisionDesc></teiHeader></istex:fulltextTEI></fulltext><metadata><istex:metadataXml wicri:clean="corpus brill-journals not found" wicri:toSee="no header"><istex:xmlDeclaration>version="1.0" encoding="UTF-8"</istex:xmlDeclaration><istex:docType PUBLIC="-//NLM//DTD Journal Publishing DTD v2.3 20070202//EN" URI="http://dtd.nlm.nih.gov/publishing/2.3/journalpublishing.dtd" name="istex:docType"></istex:docType><istex:document><article article-type="research-article" dtd-version="2.3">
<front>
<journal-meta>
<journal-id journal-id-type="e-issn">18784763</journal-id>
<journal-title>Seeing and Perceiving</journal-title>
<journal-subtitle>A Journal of Multisensory Science</journal-subtitle>
<abbrev-journal-title>SP</abbrev-journal-title>
<issn pub-type="ppub">1878-4755</issn>
<issn pub-type="epub">1878-4763</issn>
<publisher>
<publisher-name>BRILL</publisher-name>
<publisher-loc>The Netherlands</publisher-loc>
</publisher>
</journal-meta>
<article-meta>
<article-id pub-id-type="doi">10.1163/187847511X584443</article-id>
<title-group>
<article-title>Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</article-title>
</title-group>
<contrib-group>
<contrib contrib-type="author">
<name>
<surname>Sinha</surname>
<given-names>Pawan</given-names>
</name>
</contrib>
<contrib contrib-type="author">
<name>
<surname>Kalia</surname>
<given-names>Amy A.</given-names>
</name>
</contrib>
</contrib-group>
<pub-date pub-type="epub">
<year>2012</year>
</pub-date>
<volume>25</volume>
<issue>3-4</issue>
<fpage>287</fpage>
<lpage>302</lpage>
<permissions>
<copyright-statement>© Koninklijke Brill NV, Leiden, The Netherlands</copyright-statement>
<copyright-holder>Koninklijke Brill NV, Leiden, The Netherlands</copyright-holder>
</permissions>
<self-uri content-type="pdf" xlink:href="18784763_025_03-04_s004_text.pdf"></self-uri>
<kwd-group>
<kwd>haptic</kwd>
<kwd>shape acquisition</kwd>
<kwd>Tactile picture perception</kwd>
</kwd-group>
<custom-meta-wrap>
<custom-meta>
<meta-name>version</meta-name>
<meta-value>header</meta-value>
</custom-meta>
</custom-meta-wrap>
</article-meta>
</front>
<body>
<p>Seeing and Perceiving 25 (2012) 287–302 brill.nl/sp Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching? Amy A. Kalia ∗ and Pawan Sinha M.I.T., Department of Brain and Cognitive Science, 46-4089, 77 Massachusetts Avenue, Cambridge, MA 02139, USA Received 1 September 2010; accepted 17 May 2011 Abstract Numerous studies have demonstrated that sighted and blind individuals find it difficult to recognize tactile pictures of common objects. However, it is still not clear what makes recognition of tactile pictures so diffi- cult. One possibility is that observers have difficulty acquiring the global shape of the image when feeling it. Alternatively, observers may have an accurate understanding of the shape but are unable to link it to a par- ticular object representation. We, therefore, conducted two experiments to determine where tactile picture recognition goes awry. In Experiment 1, we found that recognition of tactile pictures by blindfolded sighted observers correlated with image characteristics that affect shape acquisition (symmetry and complexity). In Experiment 2, we asked drawing experts to draw what they perceived after feeling the images. We found that the experts produced three types of drawings when they could not recognize the tactile pictures: (1) draw- ings that did not look like objects (incoherent), (2) drawings that looked like incorrect objects (coherent but inaccurate) and (3) drawings that looked like the correct objects (coherent and accurate). The majority of errors seemed to result from inaccurate perception of the global shape of the image (error types 1 and 2). Our results suggest that recognition of simplistic tactile pictures of objects is largely inhibited by low-level tactile shape processing rather than high-level object recognition mechanisms. © Koninklijke Brill NV, Leiden, 2011 Keywords Tactile picture perception, haptic, shape acquisition 1. Introduction Past research demonstrates that tactile picture recognition is difficult for normally- sighted and visually-impaired individuals. Observers typically recognize only 30– 40% of test images, even though the images are usually simplistic raised-line drawings of common objects (Fig. 1). However, it is trivial for sighted observers * To whom correspondence should be addressed. E-mail: akalia@mit.edu © Koninklijke Brill NV, Leiden, 2011 DOI:10.1163/187847511X584443</p>
<p>288 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 Figure 1. Accuracy (percent correct) of sighted and blind participants in naming objects depicted in tactile or visual line drawings (data compiled from Heller, 1996; Klatzky et al ., 1993; Lederman et al ., 1990 and Loomis et al ., 1991). to recognize the same drawings visually. These findings raise the obvious question of why tactile pictures are so difficult to understand. Tactile picture recognition can be described as the combination of two processes. First, the observer must determine the shape of the stimulus. Shape acquisition is a bottom-up process requiring integration of tactile information over space and time. As observers move their hands across the image, they must comprehend and remember local pieces of information obtained by the fingertips. With each move- ment, local contours have to be aggregated into global shapes. Next, the observer must associate the shape with a particular object. This involves higher-level mech- anisms of retrieving object representations that match the perceived shape of the tactile image. Despite the numerous studies on tactile picture recognition, it is still unclear where recognition goes awry. One possibility is that observers have difficulty ac- quiring shape information from touch. Because of the limited field of view and res- olution of tactile sensory receptors, information is acquired sequentially and must be integrated during exploration. The limited field of view of touch can account for poor recognition of tactile pictures, presumably because of the increased percep- tual and cognitive demand of integration (Loomis et al. , 1991). Indeed, recognition improves when observers increase the tactile field of view by using both hands (Wi- jntjes et al ., 2008a) or multiple fingers on a hand (Klatzky et al ., 1993) to feel the image. Recognition of tactile images is generally better when they are actively rather than passively (stationary hand) scanned (Gibson, 1962; Heller, 1984; Heller and</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 289 Myers, 1983). Kinesthetic information obtained from hand movement may aid the process of integrating shape information as observers explore an image (Magee and Kennedy, 1980). Also, tactile pictures are more easily recognized when they are larger in size (Wijntjes et al ., 2008b). Interestingly, sighted and blind observers are much better (75–85% accurate) at performing tactile match-to-sample tasks with geometric shapes (Heller, 1989), suggesting that they can accurately obtain sim- plistic shape information by touch. Yet, it is still unknown just how well observers perceive the shape of more complex tactile images of objects. Alternatively, recognition of tactile pictures may be compromised during higher- level processing. It is possible that observers accurately perceive the tactile pattern, but have difficulty associating the pattern with a particular object. Heller et al . (1996) suggested that poor recognition largely results from difficulty in semantic labeling of the patterns rather than from poor tactile pattern perception. They found that recognition performance improved dramatically when participants were given categorical information (e.g., fruit, part of the body) about the tactile pictures. Another hypothesis is that recognition of tactile pictures is dependent on vi- sual experience. Tactile pictures are based on visual representations of objects, and often include pictorial cues that may be difficult to interpret by touch. Per- haps successful identification of the objects requires the use of imagery to translate the tactile inputs into a visual representation. Previous researchers have tested this possibility by comparing the performance of congenitally blind, late blind, and blindfolded sighted observers in tactile picture recognition. Heller (1989) found that congenitally blind and sighted participants performed similarly, while late- blind participants performed more accurately. However, other studies have found that congenitally blind individuals recognize more tactile pictures than sighted in- dividuals (Heller et al ., 1996; Lederman et al ., 1990). These mixed results make it unclear how much visual experience aids recognition of tactile pictures. The goal of the current study was to better understand where errors in tactile picture recognition occur. In Experiment 1, we asked normally-sighted observers to feel tactile pictures while blindfolded. We then examined recognition accuracy as a function of both low-level (e.g., image complexity) and high-level (e.g., familiarity) characteristics of the pictures. In Experiment 2, we more directly assessed percep- tion of tactile pictures by asking drawing experts, individuals with formal training in freehand drawing, to feel tactile pictures and then to draw what they felt. We were especially interested in examining drawings for trials in which the experts could not recognize the tactile picture as they were feeling them. If errors in recognition oc- cur because of difficulty in higher-level processing, like object naming, rather than in shape acquisition, then the experts should draw accurate depictions of the tactile images even if they are unable to recognize the object. Improved understanding of the process of tactile picture recognition has po- tentially important implications for the visually-impaired community. People with visual impairment use tactile graphics, typically created with Braille embossers, swell paper or tactile film, in a variety of settings, such as in education. However,</p>
<p>290 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 these graphics are often difficult for users to understand. By probing the recognition process, we can help to improve the usefulness of tactile graphics for visually- impaired individuals. 2. Experiment 1 In Experiment 1, we asked whether there are consistencies in tactile recognition across blindfolded sighted observers. In other words, are there pictures that people typically recognize, and others that they do not recognize? If so, then we were interested in characteristics of the pictures that may account for this disparity in recognition. We compared recognition of the pictures to what we broadly can call low-level traits of the tactile stimulus versus high-level traits of the objects depicted in the images. We used ratings from naïve observers and computational metrics to determine the characteristics of these pictures. Our hypothesis was that if recognition is influenced by the ease of tactile pattern perception, then recognition accuracy should correlate with low-level image char- acteristics, such as the symmetry and complexity of the image. It may be easier to acquire the shape of symmetrical images since the observer can make use of the repetition in the image; since both hands feel similar contours, there is an effec- tive increase in the field of view and decrease in the cognitive load of integrating shape information. It may also be harder to integrate shape information from more complex images since detailed information may not be resolvable by touch. If poor tactile picture recognition is the result of difficulty linking the acquired shape to an object, then higher-level properties of the objects should correlate with recognition accuracy. We looked at the correlation of recognition and two such properties — how well the depiction matched observers’ mental image of the object and the familiarity of the object. Although previous studies have indicated possible differences between blind and sighted participants in their ability to recognize tactile pictures, we chose to test sighted individuals since they were more easily accessible. We also wanted to es- tablish a baseline for performance before testing blind participants in future studies. 2.1. Method 2.1.1. Participants Twenty-two sighted individuals (11 females/11 males, mean age of 31) participated in this experiment. All participants reported having normal or corrected-to-normal vision, and no known conditions affecting tactile perception. Participants were mon- etarily compensated for their time. 2.1.2. Materials The tactile images comprised 28 raised-line drawings of common objects (Fig. 2) obtained from the Snodgrass and Vanderwart (1980) set. We selected images that were fairly simple (average rating of 2.00 where 5 is very complex). Tactile versions</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 291 Figure 2. The images of objects used as tactile pictures. of these images were produced using swell paper and a heating machine (Repro- tronics). The images were scaled to fit within an area that was approximately 16 cm wide and 11 cm tall. 2.1.3. Procedure For each trial, participants were asked to wear a blindfold and to actively explore the tactile image using both hands. We wanted participants to explore the images using natural hand movements, and without limiting their tactile field of view, thereby optimizing chances for recognition. Participants were given two minutes to name the depicted object. We found that this time constraint provided enough opportunity for participants to explore and think about the images while also ensuring that we finished the experiment in a reasonable amount of time. After they made a guess or after two minutes had expired, the experimenter removed the image. The exper- imenter recorded the subjects’ responses and the time taken to explore the image. The images were presented in a random order for each subject. The entire experi- ment took approximately one and a half hours. 2.1.4. Ratings We recruited eight naïve observers (5 females/3 males, mean age of 31) to rate the tactile images when exploring them by touch or vision. The observers first felt the images while blindfolded and rated the images on complexity (1 — very simple, 5 — very complex) and symmetry (1 — very non-symmetrical, 5 — very sym- metrical). Then the observers viewed the same images, and rated them on visual complexity and symmetry. As in Snodgrass and Vanderwart (1980), we defined complexity as the amount of detail and the intricacy of the lines in the image. We also obtained ratings for image agreement and familiarity from Snodgrass and Vanderwart (1980). Image agreement referred to how closely the depiction matched the observer’s mental image of the object (1 — low agreement, 5 —</p>
<p>292 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 high agreement). The ‘arrow’ was rated as having the lowest image agreement (2.27) and the hanger had the highest image agreement (4.73). Familiarity referred to how familiar the objects were from general experience in the real world (1 — very unfamiliar, 5 — very familiar). The ‘bell’ was rated as being the least familiar object (2.20) and the ‘key’ was rated as the most familiar object (4.85). To elimi- nate concerns that these ratings are outdated, we asked four observers to judge the images on both characteristics using the same instructions described in Snodgrass and Vanderwart (1980). The ratings of our subjects were highly correlated with the Snodgrass and Vanderwart ratings (image agreement: r = 0 . 60, p < 0 . 001; famil- iarity: r = 0 . 75, p < 0 . 001). Therefore, we used the Snodgrass and Vanderwart ratings in our analyses. 2.1.5. Computational Metrics In addition to subject ratings, we also calculated quantitative measures of symmetry and complexity for each image. Symmetry was computed by reflecting the image across several axes (0–180° in 10° increments), and then calculating the Hausdorff distance (Huttenlocher et al ., 1993) between the two halves. The Hausdorff distance is computed by first translating digitized versions of the two image halves into a set of points with x - and y -coordinates. We then computed the distance between the closest matching points in the two images. Image halves that were very similar had low Hausdorff distance scores, whereas images that were very dissimilar had high distance scores. Therefore, highly symmetric images had low Hausdorff distance scores and asymmetric images had high Hausdorff distance scores. The minimum Hausdorff distance across all possible axes was used as the symmetry metric for an image. The complexity metric reflected the average variability of edge orientations within sub-regions of the image. We reasoned that more complex images tend to have more lines, intersections between lines and curvature. To compute complexity, we ran the images through a Scharr filter (Jähne et al. , 1999), and then computed the direction of gradient changes, thereby obtaining the values of local edge ori- entations. We then computed the entropy of the distribution of orientations within a sampling window. We then averaged the entropy measure over all the sampling windows in an image to define a final complexity measure. We expect that a higher average entropy metric, as calculated by this procedure, would correspond to a more complex image and, therefore, to a lower likelihood of recognition. In other words, average entropy would be negatively correlated with image recognition. 2.2. Results On average, participants correctly recognized 40% of the twenty-eight tactile im- ages. Figure 3 shows the proportion of participants who correctly recognized each tactile picture. The tactile pictures are listed in order of increasing complexity as determined by our subject ratings. The only picture that was recognized by all participants was the ‘heart’. None of the participants were able to recognize the</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 293 Figure 3. The proportion of observers who correctly recognized each tactile picture. The images are ordered according to ratings of their complexity. picture of the ‘bread’. Also, in general, recognition of the images decreases as the complexity of the images increases, although this is not a perfect correlation (see below). 2.2.1. Recognition Accuracy vs Image Ratings We correlated recognition accuracy of each picture with the ratings provided by our naïve observers and Snodgrass and Vanderwart (1980). The ratings we obtained for visual complexity correlated highly with the ratings obtained by Snodgrass and Vanderwart (1980) ( r = 0 . 80). Also, our raters made similar judgments by vision and touch for both complexity ( r = 0 . 95) and symmetry ( r = 0 . 83). Therefore, we used the visual ratings of complexity and symmetry obtained from our observers for the following analyses. Figure 4 shows the correlations between the recognition of each tactile pic- ture and the four rating categories (symmetry, complexity, image agreement and familiarity). Both symmetry ( r = 0 . 39, p = 0 . 038) and complexity ( r = − 0 . 45, p = 0 . 017) correlated significantly with how often the tactile pictures were recog- nized. On the other hand, image agreement and the familiarity of the objects did not correlate significantly with recognition. We also tried to adjust for the range re- striction of these two variables using the Thorndike Case 2 correction (Hunter and Schmidt, 1990; Thorndike, 1949). For image agreement, the corrected correlation was less than the original ( r = 0 . 13) because our selected range of data was close to the full range. For familiarity, the corrected correlation was higher but still not significant ( r = − 0 . 29, p = 0 . 13).</p>
<p>294 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 Figure 4. Correlations between recognition accuracy (proportion of trials the tactile pictures were cor- rectly recognized) versus ratings of picture characteristics (symmetry, complexity, image agreement and familiarity). 2.2.2. Recognition Accuracy vs Computational Metrics We also correlated recognition accuracy with the quantitative measures of sym- metry (Hausdorff Distance) and complexity (Orientation Entropy). Both measures were significantly correlated with the corresponding participant ratings (symmetry: r = − 0 . 50, p = 0 . 007; complexity: r = 0 . 75, p < 0 . 001). We found that only orientation entropy was significantly correlated with recog- nition of the tactile pictures ( r = − 0 . 41, p = 0 . 031) (Fig. 5). Although Hausdorff Distance was not significantly correlated with recognition, the trend was in the ex- pected direction ( r = − 0 . 32, p = 0 . 098). 2.3. Discussion The results of Experiment 1 provide some insight into why tactile picture recog- nition is difficult. We found that some images are more easily recognized than others. Furthermore, recognition was significantly influenced by the complexity of the image, and to a lesser degree image symmetry. Image agreement (how well the depiction matched the observer’s mental image of the object) and the familiarity of the object in everyday experience seemed to have very little influence on recogni- tion. These findings suggest that tactile picture recognition is more dependent on the ease of bottom-up integration of shape information rather than high-level object matching.</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 295 Figure 5. Correlations between recognition accuracy (proportion of trials the tactile pictures were correctly recognized) and the computational metrics of symmetry and complexity. Lederman et al . (1990) also correlated recognition accuracy with various image characteristics, and found that the ease with which observers could imagine the stimulus as they were feeling it correlated highly with recognition. However, they did not find that recognition correlated significantly with the visual complexity of the images. One possible explanation is that we used a set of images with a greater range of complexity ratings than they did (visual complexity ratings of 1.00–4.25 compared to 1.2–3.25 from Snodgrass and Vanderwart, 1980). Our result is further supported by our finding that recognition accuracy was significantly correlated with a quantitative measure of image complexity. Although recognition was generally dependent on image complexity, there were a few interesting exceptions to this rule. For example, although the ‘butterfly’ and ‘kite’ images were rated as highly complex, they were also recognized very often (recognition by approximately 80% of subjects). We speculate that even though these images contained a lot of detail, the external boundaries of the image were highly detectable and distinctive. Perhaps subjects were able to recognize the im- ages based on the unique external outlines of the images. On the other hand, some- times simple images, such as the ‘spoon’, were not easily recognized. It may be that the shape of the spoon was not very distinctive, and was, therefore, easily con- fusable with other objects. The ‘uniqueness’ of shapes is likely to be an additional parameter that would help explain recognition of tactile pictures.</p>
<p>296 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 From these results, it seems that the difficulty in tactile picture recognition for this set of images is better explained by shape attributes such as complexity. Such aspects of stimulus structure are more correlated with tactile recognition perfor- mance than high-level processes such as object-indexing from memory. 3. Experiment 2 In Experiment 2, we more directly assessed observers’ perception of the shape of the tactile image by asking them to draw what they felt. To sidestep the possible confound of drawing ability, we recruited participants with significant drawing ex- perience. If recognition of tactile pictures is primarily limited by access to shape information, as suggested by Experiment 1, then the drawings should reflect these errors. However, if observers are able to accurately acquire the shape of the tactile picture, then their drawings should look similar to the original image. In these cases, we suspect recognition is limited by the difficulty in associating the shape of the im- age with a particular object. To evaluate the errors in the drawings, we asked naïve observers to rate the drawings produced when the drawing experts were unable to recognize the tactile picture. 3.1. Method 3.1.1. Participants We recruited 10 participants (6 females/4 males, mean age of 31) who were drawing experts. These individuals had formal training in freehand drawing, either through art or architecture programs. All participants were recruited from M.I.T. or the sur- rounding community and were compensated monetarily for their time. 3.1.2. Materials We used the same tactile images as in Experiment 1. 3.1.3. Procedure The procedure was the same as in Experiment 1 with the addition of the drawing task. After participants felt the tactile image for two minutes or after they guessed the name of the object, they lifted the blindfold and drew their perception of the tac- tile image. Participants were asked to make drawings even when they were unable to name the object in the tactile picture. 3.1.4. Ratings We asked 4 naïve observers to rate the drawings produced when the experts were unable to accurately name the object in the tactile image. The raters were asked if the drawing looked like a nameable object; if they said yes, then they also provided the name of the object. The raters were presented with the drawings in a randomized order. They never saw the original tactile images. 3.2. Results The expert subjects on average were able to correctly identify more tactile pic- tures (56%) than 10 age-matched ( ± 3 years) non-experts from Experiment 1 (35%)</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 297 Figure 6. Examples of the three types of drawings created by subjects in Experiment 2. In all cases, subjects were unable to name the original tactile drawing. The name of the original tactile image is noted below each drawing. ( t = 2 . 83, p = 0 . 020). As in Experiment 1, recognition accuracy was signifi- cantly correlated with ratings of symmetry ( r = 0 . 43, p = 0 . 021) and complexity ( r = − 0 . 48, p = 0 . 011), but not image agreement or familiarity. We examined the drawings produced when experts were unable to correctly name the object depicted in the tactile image. From our observations, these drawings fell into one of three categories: incoherent, coherent but inaccurate, and coherent and accurate. Figure 6 illustrates examples of these three categories of drawings. The incoherent drawings reflect errors in organizing the tactile information into global shapes; these drawings do not look like objects. The coherent but inaccu- rate drawings look like objects, but they are different from the ones presented in the original image. In this case, subjects usually made an incorrect guess about the identity of the object, which influenced their perception of the tactile stimulus. In- terestingly, in the third category of drawings, subjects drew coherent and accurate depictions of the original image even though they could not name the object when feeling the tactile image. These trials seem to reflect an error in linking the acquired shape information with a particular object. We used the ratings from the naïve observers to determine the frequency of the three types of errors. Although these frequencies are likely to be influenced by the particular images chosen for the experiment, it gives us an idea of the difficulty of shape acquisition versus object naming for this experiment. For each rater, we de- termined how often they classified drawings as not a nameable object (incoherent),</p>
<p>298 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 Figure 7. Recognition accuracy and error type for each tactile picture. The objects on the y -axis are ordered from least to most complex. This figure is published in color in the online version. as a nameable object but produced an incorrect name (coherent but inaccurate), or as a nameable object and produced a correct name (coherent and accurate). We then averaged the frequency of each type of error across raters. We found that the ma- jority of errors were due to inaccuracies in acquiring the shape of the image; on average, the raters indicated that 46% of the drawings did not even look like ob- jects and 27% of them looked like the incorrect object. For the remaining 27% of drawings, the raters were able to correctly identify the object depicted in the orig- inal tactile image, indicating that the drawers accurately obtained the shape of the tactile image. Figure 7 illustrates recognition accuracy and error type when observers did not recognize the tactile picture. The y -axis lists the objects from least complex (‘heart’) to most complex (‘butterfly’) as defined by the ratings obtained in Ex- periment 1. The x -axis indicates the proportion of subjects who either correctly recognized the tactile picture, or made one of the three errors as judged by our naïve raters. The first observation is that, generally, the simpler pictures are recog- nized more often than the more complex pictures, as was the case in Experiment 1. However, this rule does not always hold true; for example, the ‘butterfly’ was rec- ognized by all participants even though it was one of the most complex images. Secondly, most of the errors in recognition are due to inaccuracies in shape ac- quisition, especially as the images become more complex. It is also interesting to note that for some of the simplest objects, like the ‘pear’ and ‘hanger’, all of the recognition errors were due to an inability to match the shape to a particular ob- ject. Therefore, even if the shape of the image is easy to understand, it must also be distinctive enough to associate with an object representation.</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 299 3.3. Discussion The drawings produced by experts indicate that errors in tactile picture recognition occur in both low-level and high-level processes. However, the majority of errors reflect difficulty in acquiring the shape of the image. These errors could result from noisy haptic information as well as from the process of spatial and temporal in- tegration. Also, as indicated by Fig. 7, observers generally had greater difficulty integrating shape information for more complex images. Furthermore, even if subjects were able to accurately acquire the shape of the tactile image, they still were not always able to name the object. One possible expla- nation for this behavior is that, for these trials, subjects do not have a representation of the global shape until they draw the image. Alternatively, subjects may have an accurate understanding of the global shape, but it is not distinct enough to be associated with a particular object. 4. General Discussion The goal of this study was to better understand why tactile picture recognition is so difficult. We hypothesized that difficulty in recognition may arise at two possible places in the process. First, observers may have trouble integrating global shape information from local samples of tactile information. Secondly, once observers acquire the shape of the tactile image, they may have difficulty associating it with a particular object. The results of these experiments suggest that for simple raised-line drawings of common objects, recognition is largely compromised by shape acquisition. Touch inherently has lower resolution and field of view as compared to vision. As ob- server’s move their fingers over an image, they have to retain local pieces of tactile input so that it can be integrated with incoming input. Hence, it is possible that er- rors arise from the process of integrating local pieces of information into a global structure. Noisy motor control and poor spatial localization of the hand might also contribute to the error in integration. The integration process also requires more cognitive demands since local pieces of information must be retained in memory. All these factors are likely to contribute to the difficulty of tactile shape acquisition. We did find that shape acquisition, and, therefore, recognition is significantly correlated with the complexity of the stimuli. Simple, geometric patterns like the ‘crescent moon’ and ‘heart’ were easily recognized; it was also easy for observers to acquire the overall shape of image. It may be more difficult to acquire the shape of complex images for several reasons. When the finger pad overlaps multiple lines, interference due to poor resolution can make it difficult to parse the different seg- ments and their orientations (Loomis, 1981). Also, it may be more confusing to trace the image if there are multiple intersections between segments. Interestingly, one exception to the link between recognition and image complexity was the ‘but- terfly’ image. Although this picture ranked as the most complex image in our set, observers were able to easily recognize it. Perhaps observers found it easy to trace</p>
<p>300 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 Figure 8. Blurred and thresholded versions of our stimulus set arranged in decreasing order of tactile recognizability. The most readily recognizable stimuli in the tactile condition appear to be especially robust against diffusive degradations of the kind used here. the outer boundary of the image unlike some of the other complex images. Further- more, the distinctiveness of the butterfly shape may have facilitated recognition. It is interesting to note that a shape’s tactile recognizability appears to be corre- lated with how well it can tolerate degradation. Figure 8 illustrates this point. The images were generated by blurring and thresholding our original stimuli shown in Fig. 2. The individual items are arranged in raster order of tactile recognizability (data from Experiment 1). To a first approximation, the first several images appear to maintain their recognizability better than the last several. It is also interesting to note the exceptions to this general idea. For instance, the cup appears to be rec- ognizable after degradation, but is not well recognized in the tactile condition. In future work, it will be interesting to investigate how reasonable it is to approximate tactile estimates of shape as diffusions of the original curves and the associated obliteration and blending of detail. Although the majority of recognition errors seem to be due to shape acquisition, observers sometimes were unable to recognize the image even if they understood its shape. In Experiment 2, these errors were evident when observers were able to accurately draw the tactile picture, even though they could not recognize it when feeling it. This type of error has been reported in the past, and the frequency of this error in Experiment 2 was consistent with what has been found previously (Wijn- tjes et al ., 2008a). It may be that for these cases observers are unable to visualize the global shape of the image until they draw it. The act of drawing forces the in- tegration of local information, and allows the observer to visualize all the parts of the shape simultaneously. It may also be that the drawing is less noisy than the ob- servers’ mental image of the shape, which also allows for easier recognition. This may appear counterintuitive, but is based on the possibility that motoric reproduc- tion of a felt contour might be more accurate than an internal visualization of the same, because the hand acts as both the sensor and the generator in the former case. On a side note, we interestingly found that drawing experts were better at tactile picture recognition than non-experts. It may be that formal training in drawings</p>
<p>A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 301 aids recognition in several ways. Drawing experts may have better motor control, and, therefore, are better able to acquire the shape of the stimulus. Alternatively, their expertise in producing two-dimensional depictions of real world objects may also aid recognition. Although this comparison was not the focus of our study, this finding brings up interesting questions for future investigation. In conclusion, our results suggest that integration of shape information is a sig- nificant obstacle to tactile picture recognition. Observers appear to have difficulty integrating local samples of the tactile stimulus into a global whole. From the ap- plied perspective, the improvement of tactile picture perception for sighted as well as blind individuals will require facilitation of this integration process. This might be accomplished by novel training approaches that can help improve integration skills, or by the development of new styles of tactile depictions that mitigate inte- grative requirements. Acknowledgements This work was supported by a postdoctoral fellowship from Fight for Sight and a grant from the National Eye Institute (F32EY019622) to Amy Kalia and a grant from the James McDonnell Foundation to Pawan Sinha. We would also like to thank the M.I.T. Disability Services Office for their assistance in creating the tactile images. References Gibson, J. J. (1962). Observations on active touch, Psycholog. Rev. 69 , 477–491. Heller, M. A. (1984). Active and passive touch: the influence of exploration time, J. Gen. Psychol. 110 , 243–249. Heller, M. A. (1989). Picture and pattern perception in the sighted and the blind: the advantage of the late blind, Perception 18 , 379–389. Heller, M. A. and Meyers, D. S. (1983). Active and passive tactual recognition of form, J. Gen. Psy- chol. 108 , 225–229. Heller, M. A., Calcaterra, J. A., Burson, L. L. and Tyler, L. A. (1996). Tactual picture identification by blind and sighted people: effects of providing categorical information, Percept. Psychophys. 58 , 310–323. Hunter, J. E. and Schmidt, F. L. (1990). Methods of Meta-Analysis: Correcting Error and Bias in Research Findings . Sage, Newbury Park, CA, USA. Huttenlocher, D. P., Klanderman, G. A. and Rucklidge, W. J. (1993). Comparing images using the Hausdorff distance, IEEE Trans. Patt. Anal. Machine Intell. 15 , 850–863. Jähne, B., Scharr, H. and Körkel, S. (1999). Principles of filter design, in: Handbook of Computer Vision and Applications , Vol. 2, Jähne, B., Haußecker, H. and Geißler, P. (Eds), pp. 125–151. Academic Press, San Diego. Klatzky, R. L., Loomis, J. M., Lederman, S. J., Wake, H. and Fujita, N. (1993). Haptic identification of objects and their depictions, Percept. Psychohys . 54 , 170–178.</p>
<p>302 A. A. Kalia, P. Sinha / Seeing and Perceiving 25 (2012) 287–302 Lederman, S. J., Klatzky, R. L., Chataway, C. and Summers, C. D. (1990). Visual mediation and the haptic recognition of two-dimensional pictures of common objects, Percept. Psychophys. 47 , 54–64. Loomis, J. M. (1981). Tactile pattern perception, Perception 10 , 5–27. Loomis, J. M., Klatzky, R. L. and Lederman, S. J. (1991). Similarity of tactual and visual picture recognition with limited field of view, Perception 20 , 167–177. Magee, L. E. and Kennedy, J. M. (1980). Exploring pictures tactually, Nature 283 , 287–288. Snodgrass, J. G. and Vanderwart, M. (1980). A standardized set of 260 pictures: norms for name agree- ment, image agreement, familiarity, and visual complexity, J. Exper. Psychol.: Human Learning and Memory 6 , 174–215. Thorndike, R. L. (1949). Personnel Selection . Wiley, New York, USA. Wijntjes, M. W. A., van Lienen, T., Verstijnen, I. M. and Kappers, A. M. L. (2008a). Look what I have felt: unidentified haptic line drawings are identified after sketching, Acta Psychol . 128 , 255–263. Wijntjes, M. W. A., van Lienen, T., Verstijnen, I. M. and Kappers, A. M. L. (2008b). The influence of picture size on recognition and exploratory behavior in raised-line drawings, Perception 37 , 602–614.</p>
</body></article></istex:document></istex:metadataXml><mods version="3.6"><titleInfo><title>Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title></titleInfo><titleInfo type="alternative" contentType="CDATA"><title>Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?</title></titleInfo><name type="personal"><namePart type="given">Pawan</namePart><namePart type="family">Sinha</namePart><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Amy A.</namePart><namePart type="family">Kalia</namePart><role><roleTerm type="text">author</roleTerm></role></name><typeOfResource>text</typeOfResource><genre type="research-article" displayLabel="research-article"></genre><originInfo><publisher>BRILL</publisher><place><placeTerm type="text">The Netherlands</placeTerm></place><dateIssued encoding="w3cdtf">2012</dateIssued><dateCreated encoding="w3cdtf">2012</dateCreated></originInfo><language><languageTerm type="code" authority="iso639-2b">eng</languageTerm><languageTerm type="code" authority="rfc3066">en</languageTerm></language><physicalDescription><internetMediaType>text/html</internetMediaType></physicalDescription><subject><genre>Keywords</genre><topic>haptic</topic><topic>shape acquisition</topic><topic>Tactile picture perception</topic></subject><relatedItem type="host"><titleInfo><title>Seeing and Perceiving</title><subTitle>A Journal of Multisensory Science</subTitle></titleInfo><titleInfo type="abbreviated"><title>SP</title></titleInfo><genre type="Journal">journal</genre><identifier type="ISSN">1878-4755</identifier><identifier type="eISSN">1878-4763</identifier><part><date>2012</date><detail type="volume"><caption>vol.</caption><number>25</number></detail><detail type="issue"><caption>no.</caption><number>3-4</number></detail><extent unit="pages"><start>287</start><end>302</end></extent></part></relatedItem><identifier type="istex">116341E5C4AB855FACAAB4DBF5D84AA2479A0825</identifier><identifier type="DOI">10.1163/187847511X584443</identifier><identifier type="href">18784763_025_03-04_s004_text.pdf</identifier><accessCondition type="use and reproduction" contentType="copyright">© Koninklijke Brill NV, Leiden, The Netherlands</accessCondition><recordInfo><recordContentSource>BRILL Journals</recordContentSource><recordOrigin>Koninklijke Brill NV, Leiden, The Netherlands</recordOrigin></recordInfo></mods></metadata><serie></serie></istex></record>Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Ticri/CIDE/explor/HapticV1/Data/Istex/Corpus
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 004158 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Istex/Corpus/biblio.hfd -nk 004158 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien
|wiki= Ticri/CIDE
|area= HapticV1
|flux= Istex
|étape= Corpus
|type= RBID
|clé= ISTEX:116341E5C4AB855FACAAB4DBF5D84AA2479A0825
|texte= Tactile Picture Recognition: Errors Are in Shape Acquisition or Object Matching?
}}
| This area was generated with Dilib version V0.6.23. |  |