Disparity map DBSCAN clustering-based region of interest extraction method

Abstract

The invention discloses a disparity map DBSCAN clustering-based region of interest extraction method, which comprises the following steps of 1, performing matching calculation on a binocular view to obtain a disparity image; 2, deforming the parallax image into an image matrix with one row and N column vectors, wherein N is the product of the width W and the height H of the image; 3, taking the absolute value distance of the parallax as a DBSCAN sample distance, and carrying out the clustering analysis of the image matrix through employing a DBSCAN clustering algorithm; and 4, mapping a clustering result back to a certain path in the original binocular view, and obtaining a region of interest according to the clustering result. The method does not need manual interaction and sample set training, can automatically extract the region of interest at each depth, is not affected by a complex background, and facilitates the further analysis and processing of the image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for extracting regions of interest based on disparity map DBSCAN clustering. Background technique [0002] Region of Interest (ROI) refers to an image region selected in some form from the processed image, and this region is the focus of subsequent image analysis. Generally speaking, the region of interest is the region where the target object is located in the image, rather than the background region. The region of interest extraction technology can reduce image processing time and improve processing accuracy. [0003] Existing region of interest extraction is mostly based on monocular images, usually using manual interaction or image threshold segmentation method to find the corresponding target region and regard it as the region of interest, its accuracy is greatly affected by the image itself, such as uneven illumination, Small differences between the fore...

AI Technical Summary

Problems solved by technology

[0003] Existing region of interest extraction is mostly based on monocular images, usually using manual interaction or image threshold segmentation method to find the corresponding target region and regard it as the region of interest, its accuracy is greatly affected by the image itself, such as uneven illumination, Small differences between the foreground and the background will cause difficulties in extraction

Target detection technology can also be used as a means of extracting regions of interest, and the target frame of the detection output is used as the region of interest for subsequent processing, but it requires a large number of samples as offline training data, which is difficult for occasions where sample acquisition and labeling costs are high. Be applicable

Images