Object Recognition System Design in Computer Vision: a Universal Approach
This work addresses object detection tasks for computer vision applications, but it appears incremental as it builds on existing methods like LibSVM with specific improvements.
The paper tackles object recognition in computer vision by proposing a universal system that includes a novel texture descriptor (Gray Level-Radius Co-occurrence Matrix), automatic feature synthesis, and a faster SVM parameter optimization method, achieving a 20-100 times speedup in optimization.
The first contribution of this paper is architecture of a multipurpose system, which delegates a range of object detection tasks to a classifier, applied in special grid positions of the tested image. The second contribution is Gray Level-Radius Co-occurrence Matrix, which describes local image texture and topology and, unlike common second order statistics methods, is robust to image resolution. The third contribution is a parametrically controlled automatic synthesis of unlimited number of numerical features for classification. The fourth contribution is a method of optimizing parameters C and gamma in LibSVM-based classifier, which is 20-100 times faster than the commonly applied method. The work is essentially experimental, with demonstration of various methods for definition of objects of interest in images and video sequences.