A Stereo Image Segmentation Method Based on Convolutional Neural Network

Abstract

The embodiment of the present invention discloses a stereoscopic image segmentation method based on a convolutional neural network, including transferring the left segmented image to the right segmented image through coherent parallax pixels in the stereoscopic disparity image; training the convolutional neural network to extract the left The features in the view and the corresponding left segmentation map are the training set. After training, the features in the right view are extracted as input to generate the target probability map of the right view. According to the target probability map, credible pixels are extracted; according to the coherence Migrating pixels in the parallax segmentation map and trusted pixels in the target probability map, synthesizing the right view segmentation initialization sampling points; using the energy optimization model to solve the right view segmentation map, and finally binarizing the right view segmentation map to obtain The final right view segmentation result. The implementation of the present invention prevents the stereoscopic image segmentation quality from being overly dependent on the accuracy of the stereoscopic disparity map, and at the same time uses the convolutional neural network to solve the problem of poor segmentation effect of the occluded area of ​​the stereoscopic image, and obtains a high-quality stereoscopic image segmentation effect.

Description

technical field [0001] The invention belongs to the technical field of image processing, and specifically refers to a three-dimensional image segmentation method based on a convolutional neural network. Background technique [0002] Image segmentation is the technology and process of dividing an image into several specific regions with unique properties and extracting objects of interest. In the prior art, great success has been achieved for single image segmentation, but left-right consistent stereo image segmentation is still a very challenging problem. [0003] As stereoscopic images and videos have become more and more popular in recent years, the demand for stereoscopic image and video editing has gradually heated up. Stereoscopic image segmentation is a challenging and hot topic in the field of stereoscopic image and video editing. Compared with single image segmentation, stereoscopic image segmentation puts forward more requirements. The object of stereoscopic image...

AI Technical Summary

Problems solved by technology

[0010] The technical problem to be solved by the embodiments of the present invention is to provide a stereoscopic image segmentation method based on a convolutional neural network. According to the left and right views of the stereoscopic image and the segmented left view, the method uses the coherent parallax pixels in the stereoscopic parallax map Point, migrate the left segmentation map to the right segmentation map to avoid the stereo image segmentation quality overly dependent on the accuracy of the stereo disparity map, construct a convolutional neural network, generate the right view target probability map, and extract advanced features of the image to deal with stereo image occlusion The segmentation problem of the region solves the problem of poor segmentation effect of the occluded region of the stereoscopic image. According to the coherent disparity segmentation map migration pixels and the credible pixels in the target probability map, the right view segmentation initialization sampling points are synthesized, and a linear energy minimum is designed. Optimize the model, solve the right view segmentation map, and obtain high-quality stereoscopic image segmentation effect

Images