An image semantic segmentation method based on local region conditional random field model

Abstract

The invention relates to an image semantic segmentation method based on a local area conditional random field model. A full convolution neural network structure of the invention extracts input picturefeatures and obtains a rough segmentation result. The region selection structure filters the edge of the segmentation result map, and selects the segmentation result as the largest circumscribed rectangle of the pedestrian, bicycle and motor vehicle parts. The local region conditional random field model establishes the conditional random field model in the rectangular region and refines the segmentation result of the rectangular region. The invention effectively combines the advantages of the precision of the conditional random field model with the advantages of the speed of the full convolution neural network. The computational method of the conditional random field model is optimized so that the time complexity of the model is greatly reduced. The segmentation accuracy of the traditional full convolution neural network is improved. The application of probability graph model and full convolution neural network is designed as an end-to-end system.

Description

technical field [0001] The invention belongs to the technical field of computer vision and relates to an image semantic segmentation method based on a local area conditional random field model. Background technique [0002] In the past two decades, deep convolutional neural networks have gradually become a powerful tool for image understanding in computer vision. Recently, Convolutional Neural Networks have shown good results on the task of semantic segmentation of images. As a cornerstone technology of image understanding, image semantic segmentation plays a pivotal role in many aspects, such as autonomous driving, drone applications, wearable devices, etc. How to design a segmentation algorithm that can balance the accuracy of semantic segmentation network and the speed of semantic segmentation has become the mainstream of current research. [0003] At present, the semantic segmentation application of images has gradually developed into two main directions according to t...

AI Technical Summary

Problems solved by technology

However, the process of mean field inference in conditional random field models is similar to the iterative application of bilateral filters, which ultimately limits the application of this method in real-time systems due to the high computational complexity of the underlying bilateral filtering step.

Images