New Effective Method for Object Recognition based on Probabilistic Pruning of Visual Characteristics in HMAX

The human visual system (HVS) recognizes object in the crowded scenes with high speed and accuracy. So far, many object recognition models based on HVS, like HMAX, have been developed. In this paper, the new effective method based on HMAX is proposed called Probabilistic Selection HMAX (PSHMAX). HMAX main problem is random patch extraction which extracts two useless patches. First, patches involving low information that cause more computational complexity with no useful result. Second, patches with wrong information from background that produce wrong output. In the proposed method, the optimum patches involving maximum useful information are extracted in the random way which has two steps: first is producing poll of patches involving maximum information, second is patches extracting with useful information from poll. To evaluate the proposed method, we apply it to object categorization and conduct experiment on the Caltech5 and Caltech101 databases. Results demonstrate that the proposed method has a higher performance than the HMAX and existing architectures having a similar framework.