Image pixel labeling method based on super pixel level features

Abstract

The invention discloses a method, for labeling object categories which pixels belong to, based on super pixel level features and belongs to the field of scene understanding and image semantic segmentation. Labeled training set images are used for training, and a to-be-labeled image is labeled by using the steps of firstly, extracting and encoding the features of each pixel; secondly, acquiring super pixels through 'over-segmentation; building different-level neighborhood areas for each super pixel according to different distances; for each area, using Max-Pooling comprehensive features to connect the features of different levels so as to obtain the level features of the super pixels; thirdly, using the level features to build and solve random field models based on the super pixels through linear SVM to obtain image pixel labels. By the method, pixel labelling time can be reduced, and labeling precision can be increased. In addition, the CUDA parallel calculation technology can be used during labeling so as to further increase labeling speed.

Description

technical field [0001] The invention relates to labeling the category of objects to which each pixel of an image belongs, and can realize object detection, segmentation and recognition in the same process, and is the basis of image scene understanding. Background technique [0002] At present, the pixel labeling technology mainly uses a conditional random field (CRF) model to label each pixel in the image one by one. The label of each pixel is represented by a random variable. Its joint probability distribution is determined by the unary potential function (referred to as unary term) and the binary potential function (referred to as binary term). The unary term for each pixel is usually determined by the probability of the pixel label output by the pixel classifier. Since the amount of calculation of the classifier is usually large and the number of image pixels is large, it is very time-consuming to calculate the unary term for each pixel. A binary term is usually determ...

AI Technical Summary

Problems solved by technology

However, since this method uses a nonlinear SVM to calculate the unary term, the speed of labeling is still slow due to the large amount of calculation of the nonlinear kernel function.

In addition, since this method only uses the histogram to synthesize the features of the adjacent superpixels to form a single-layer region as the features of the superpixels, the labeling accuracy also needs to be improved.

Images