Foreground object image segmentation method based on deep convolutional neural network

Abstract

The invention discloses a foreground object image segmentation method based on a deep convolutional neural network, sets segmentation tasks as a dense marking problem, and provides a pixel-level imagesegmentation model FOSeg model based on the deep convolutional neural network. The FOSeg model is a foreground object image segmentation model which can predict the possibility that each pixel is a foreground object and supports end-to-end training. The FOSeg model segmentation flow chart is shown in Figure 1. First, an original image is input, convolution, convergence, linear rectification and other operations of the deep convolutional neural network are carried out; the method comprises the following steps: obtaining a feature mapping score graph by using a bilinear interpolation algorithm,inputting the feature mapping score graph into a shunt aggregation module, obtaining a rough segmentation graph by using an up-sampling operation of the bilinear interpolation algorithm, and finallysending the obtained rough segmentation graph into a conditional random field model to further refine a segmentation result, thereby obtaining a fine segmentation image.

Description

【Technical field】 [0001] The invention relates to the technical field of computer vision, in particular to a foreground object image segmentation method based on a deep convolutional neural network. 【Background technique】 [0002] Foreground object segmentation models can be divided into two categories according to whether the segmentation is category-dependent or not. (1) Class-independent segmentation: No matter how many classes the object in the image has, a segmentation model that only extracts the mask of the foreground object; (2) Class-specific segmentation: Learn from class-labeled data and seek to segment Class-specific segmentation models for . [0003] (1) Class-independent segmentation [0004] According to the different segmentation strategies and purposes, the segmentation that does not depend on categories is divided into the following three types: [0005] Interactive image segmentation model: such as the GrabCut [10] model is to let people use borders or ...

AI Technical Summary

Problems solved by technology

Compared with the existing better image region segmentation model based on depth features, it has higher accuracy and better solves the problem of decreased positioning accuracy due to the invariance of DCNNs

Images