Image Classification Based On Pcanet And Salient Feature Fusion

Abstract

Aiming at the shortcomings of traditional image classification model in extracting features, we propose an improved color contrast algorithm to extract higher quality saliency map. We first analyze the feature extraction ability of HC saliency algorithm in image classification and improve it by adding the location information, then we propose a novel features fusion module to combine the saliency map with the output features from PCANet to enhance the feature expression, contributing to classification capability of the model. The accuracy on Caltech101 and Pascal VOC2007 can achieve excellent performance by using our method.