Abstract:Traditional pattern recognition methods mainly focus on ‘classification’ rather than ‘cognition’. A new object which has been never seen before is often falsely classified as a certain kind of category studied, while humans will just respond unknown. This is mainly due to the fact that the human recognition process is intimately associated with the human memory system. So far, most memory models are studied in word list and few are reported in visual images. In this paper, we incorporate the human memory into visual image classification and propose a memory model for visual image storage and recall based on Bayesian decision (VISRBD). All the studied images are stored in the form of feature vector in the memory space. And each image feature component is correctly copied with certain probability generated by an exponential distribution. When a probe image is presented, the likelihood ratio between it and each studied image is calculated based on probabilistic theory. Then the odd in favor of an old over a new item is computed based on the ratio values between the test image and all the studied images. According to the odd value, the Bayesian decision rule for image classification is performed. Experimental results show that the proposed VISRBD model can gain good classification performance and the false alarm rate is far lower than SVM.

A Memory Model for Visual Image Storage and Recall Based on Bayesian Decision