Camouflage target image segmentation method based on information mining

Abstract

The invention belongs to the technical field of scene segmentation in computer vision, and discloses a camouflage target image segmentation method based on information mining. The PFNet sequentially comprises a multi-layer feature extractor, a positioning module and a focusing module, wherein the multi-layer feature extractor obtains context features of different levels by using a traditional feature extraction network; the positioning module firstly uses RGB feature information to preliminarily determine the position of a camouflage target in an image; and the focusing module is used for mining information and removing interference information on the basis of the RGB feature information and the preliminary position information of the image, and finally determining the boundary of the camouflage target step by step. According to the method, the concept of interference information is introduced into a camouflage target segmentation problem, a new information exploration and interference information removal strategy is developed, and segmentation of a camouflage target image is facilitated. From the perspective of the result, the PFNet segmentation result is very excellent, and the fine degree of the camouflage target boundary is also satisfactory. Meanwhile, the method is wider in applicability.

Description

technical field [0001] The invention belongs to the technical field of scene segmentation (SceneSegmentation) in computer vision, the realization result is the segmentation of image content, and particularly relates to a segmentation method of a camouflage target in a real environment image. Background technique [0002] Two-dimensional image segmentation refers to the technology of distinguishing the pixels belonging to different objects in the image to determine the size, shape and position of the target in the environmental image. It is a key step from image processing to image analysis and has great application value. . In recent years, methods related to scene segmentation, such as object detection, depth estimation, salient region detection, and shadow detection, have achieved significant performance improvements. [0003] There are quite a few creatures in nature that have evolved superb camouflage skills and can camouflage themselves to blend into the surrounding en...

AI Technical Summary

Problems solved by technology

This method mainly utilizes the discontinuity of depth and content between the foreground and background of such regions, however, for camouflaged targets, the difference in these aspects is not obvious, and the structure of camouflaged targets is usually more complex

Therefore, the SRS method cannot solve the problem of camouflaged target segmentation very well.

Images