Research On The Detection Method Of The Occlusion Boundary Of The Object Image

At present,2D images is one of the most widely used forms of visual data,which is of great significance to the dissemination and display of visual information.The research tasks in the field of computer vision mostly depend on the data form of 2D images.Although 3D images technology has made great progress in recent years,due to the high cost of 3D data acquisition,the resource consumption is also relatively large,in the foreseeable future,the data form of 3D images is still difficult to compare with 2D images.Although the two-dimensional image data is simple and easy to store,the lack of depth information in the world results in the lack of information in the twodimensional image.Occlusion is a common phenomenon in images,which is the direct expression of image information loss.This article focuses on the occlusion phenomenon in general 2D images.To solve the problem of occlusion boundary detection in two-dimensional image.In this paper,an end-to-end occlusion edge detection method based on the encoder-decoder is proposed,which can detect the occlusion between objects according to the occlusion phenomenon in images,and judge the occlusion relationship between objects through the left-hand rule.In view of the imbalance between boundary pixels and non-boundary pixels in 2D images,a new loss function is proposed to balance the weights of boundary pixels and non-boundary pixels,and the experimental results show that the loss function can effectively reduce the error rate of false occluded boundary pixels.The method in this paper has carried out systematic verification experiments on public datasets BSDS500 and PIOD.Through quantitative and qualitative experimental analysis,we prove the proposed method can accurately detect the occlusion boundary in the object image,and effectively alleviate the problem of low occlusion boundary recognition efficiency caused by the imbalance between boundary pixels and nonboundary pixels..