Image segmentation method and device

Abstract

The invention provides an image segmentation method and device. For a target object, determining a plurality of corresponding label fusion strategies according to labels in the sample image, the category number m of the labels being greater than or equal to 3, the labels not including a background, and the number n of the label fusion strategies being less than or equal to m; training an image segmentation model for each label fusion strategy to obtain a segmentation result output by each image segmentation model; and fusing the segmentation results output by the image segmentation models to obtain the target object. According to the method, structural correlation between different labels is considered, and recognition of the target object is met through image segmentation based on label fusion. The segmentation effect on the target object is obviously superior to the situation that each model identifies one of the plurality of labels, and is also obviously superior to the situation that the plurality of labels are identified through a single model.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a multi-category image segmentation technology. Background technique [0002] When processing image segmentation containing multi-category labels, the background (label) is not considered, and the number of categories of labels is m≥3. The processing for each label is generally divided into the following two methods: [0003] Method 1. Train an image segmentation model whose output is m target objects. [0004] Method 2. Train m image segmentation models, each image segmentation model is aimed at a target object, and the results of the m image segmentation models are fused to obtain m target objects. [0005] In method 1, one image segmentation model handles all categories, which will lead to difficulty in parameter adjustment, and the segmentation effect between categories cannot be balanced; method 2 will obviously increase the workload of image segmentation model tra...

AI Technical Summary

Problems solved by technology

[0005] In method 1, one image segmentation model handles all categories, which will lead to difficulty in parameter adjustment, and the segmentation effect between various categories cannot be balanced; method 2 will obviously increase the image Splitting the workload of model training is inefficient, and it is difficult to apply multiple models

Moreover, neither of these two approaches takes into account the structural correlations that exist between the targets to be segmented

Images