Image retrieval remains a difficult task, in spite of the many research efforts applied over the past decade or more, from IBM's QBIC onwards. Colour, texture and shape have all been used for content-based image retrieval (CBIR); texture is particularly effective, alone or with colour. Many researchers have expressed the hope that textures can be organised and classified in the way that colour can; however, it seems likely that such an ambition is unrealisable. While the goal of content-based retrieval is to retrieve "more images like this one," there is the difficulty of judging what is meant by similarity for images. It seems appropriate to search on what the images actually look like to potential users of such systems. No single computational method for textural classification matches human perceptual similarity judgements. However, since different methods are effective for different kinds of textures, a way of identifying or grouping such classes should lead to more effective retrievals. In this research, working with the Brodatz texture images, participants were asked to select up to four other textures which they considered similar to each of the Brodatz textures in turn. A principal components analysis was performed upon the correlations between their rankings, which was then used to derive a 'mental map' of the composite similarity ranking for each texture. These similarity measures can be considered as a matrix of distances in similarity space; hierarchical cluster analysis produces a perceptually appropriate dendrogram with eight distinct clusters.
Texture plays an important part in many Content Based Image Retrieval systems. This paper describes the results from a human study, which asked 30 volunteers to classify images from the Brodatz Textures album. We use these results to derive a subset which show good agreement among the different individuals. The results for this subset were used to evaluate the retrieval performance of a range of statistical, Fourier- based, and spatial/spatial filtering methods. However, no one computational method works well for all textures, unlike the human visual system. We show how each of the ten methods correlates with the rankings from the human studies. The results typically match for only about 20% - 25% of the images. Combining two techniques can improve the retrieval performance, as judged by human users. We also identify a further subset of the Brodatz images where no computer method correlates significantly with the composite human ranking. Of the 85 images selected by the human study, only 64 have any significant correlation with one or more of the computational methods in this paper. The excluded images, where human users agree with each other, but none of the methods we evaluated did, provide a further challenge to texture-based image retrieval techniques.
Content-based Image Retrieval is an area of growing interest. Various approaches exist which use color, texture, and shape for retrieving 'similar' images from a database. However, what do we mean by 'similar'. Traditionally, similarity is interpreted as distance in feature space. But this does not necessarily match the human users' expectations. We report on two human studies, which asked volunteers to select which imags they considered to be 'most like' each image from the Brodatz dataset. Although the images from the Brodatz set have the advantage of being an agreed standard in texture analysis, Brodatz certainly did not select his images with this in mind. The results from this study provide a justification for selecting a subset of the Brodatz data set for use in evaluating texture-based retrieval techniques. Images which humans have difficulty in agreeing which other images are 'most like' are also poor choices for comparison. Our result indicate which images are most likely to be classified as 'similar' by individual humans and that can also serve to evaluate computer-based retrieval techniques.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.