| Recent advances in digital imaging and data exchange have revolutionized the way visual media is created and distributed. The problem of search, recognition, and classification of visual media is overwhelming for individual user and has stimulated the development of automated systems that can handle these tasks efficiently. Humans recognize objects and shapes in their surroundings with ease and efficiency that current state-of-the-art systems do not approach. As observers, we are able to identify objects under a variety of conditions such as changes in lighting, the presence of clutter and other elements disturbing the visual field. Moreover, observers recognize the same object from many different views (from various angles, up close, from afar, etc...).;The apparent ease with which humans recognize objects serves as motivation for developing a biologically inspired model for texture and shape recognition which mimics particular behaviors of human visual processing that aid in object recognition. It is widely believed that the visual system extracts relevant features in the frequency domain, and to this end we propose a feature extractor which employs a discrete Fourier transform in the polar space.;This feature is shown to be efficient in two very important texture analysis problems, namely, texture classification and texture segmentation. We test the algorithm using several benchmarks from the Outex database, a unified framework designed for the empirical evaluation of texture analysis algorithms. This database contains a large collection of synthetic and natural images which require the recognition of various naturally-occurring transformations, such as rotation, scaling, and translation. Our system improves on the state-of-the-art.;We further investigate the robustness of the proposed algorithm by applying it to the results of image segmentation in order to capture the characteristics of the boundary contour, recognizing its shape. The databases used are the Kimia data sets and the MPEG-7 CE-1 Shapes database which contain different classes of object silhouettes that are rotated, scaled, occluded and partially represented. Even on these problems, our system achieves results comparable to state-of-the-art shape recognition algorithms.;To the best of our knowledge, the algorithm we propose is the first algorithm that can achieve state-of-the-art results for both texture and shape recognition. |
