Biologically Inspired Hierarchical Model for Feature Extraction and Localization

Some of the most important problems of computer vision are feature extraction and subsequent localization of those features in a new image. Since it is computationally prohibitive to search for the features over all possible locations and scales, it is necessary to design an algorithm that can selectively focus on and process information from only some regions within the image. In this work we present such an algorithm that is biologically inspired and performs a hierarchical search, from coarse to fine, in order to minimize the computational costs. The algorithm is very robust to non-linear image transformations such as changes in scale, rotation, skew, addition of noise, and changes in brightness and contrast. We demonstrate the computational efficiency as well as the effectiveness of the algorithm on several real world images