Image Segmentation Algorithm Based on Jump Feature Fusion and Rich Features

Abstract

With the development of deep learning, convolution neural networks have become the mainstream of computer vision algorithms. In recent years, the biggest problem of applying convolution neural network to image segmentation is that it can not achieve accurate segmentation at the last layer, and it will cause resolution loss when extracting features. In order to solve these two problems, we add jump feature fusion methods after Entry, Middle, ExitFlow and ASPP module respectively, so that the feature loss will not be serious when extracting features. In the process of feature restoration, a module combining bilinear upsampling and deconvolution is added to further enrich the feature graph and make the features robust. The experimental results show that the results exceed the performance of other previous algorithms. We demonstrate the effectiveness of the proposed model on PASCAL VOC 2012, achieving the test set performance of 85.5%.