Region of interest extraction method of pixel level

Abstract

The invention discloses a region of interest extraction method of a pixel level. The method is used to reduce a data scale and improve image processing efficiency. The method comprises the following steps: using visual attention models to obtain a saliency map and carrying out saliency map binarization; after a salient point is obtained, clustering, and optimizing a problem generated during the clustering; simultaneously, carrying out original image binarization and extracting a binary image outline of the original image through scanning the binary image; taking the optimized clustering point as a seed point to perform filling; performing mask with the original image and then extracting the region of interest of the image. In the invention, the saliency map is taken as a base; the pixel point which is the salient point in the saliency map is taken as a target. An operation speed of extracting the region of interest in the image can form a linear relation with a number of the salient point. A leakage point rate of the region of interest at least decreased to a half of the original leakage point rate and a classification error rate does not increase obviously.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and relates to a method for extracting a region of interest at the pixel level. Background technique [0002] Psychological research has found that what attracts people's attention is the areas in the image that can generate novel stimuli or those areas in the image that correspond to the stimuli that people expect. These areas are the areas of interest, also known as visual salience. regions, which contain most of the image information. This makes it possible to obtain most of the information in the image if the region of interest in the image is obtained when processing the image, and focusing on analyzing these regions can reduce the amount of calculation and thus cope with the large increase in the amount of image data . At the same time, more accurate methods of region-of-interest extraction and analysis are needed to accurately identify and operate relevant regions in images, so...

AI Technical Summary

Problems solved by technology

[0009] The saliency map obtained by using Itti-Koch will become blurred; the Stentiford visual attention model calculates the global saliency of the corresponding region of the input image, and evaluates the overall saliency of the image region, so it is difficult to extract low-saliency objects using this model area of ​​interest

In addition, the resolution of the saliency map generated by this model is the same as that of the input image, but there are more noise points. When using the saliency map, it is necessary to pay attention to the error effect of the noise; if the spectral residual model is used, the accuracy loss of the saliency map will be serious.

The Hu-Rajan-Chia model is also a purely computational visual saliency model, but for images with inconspicuous texture features, its saliency calculation results still need to be strengthened

In addition, the resolution of the saliency map extracted by the model is also lower than that of the input image

[0010] Therefore, each visual attention model has its own advantages and disadvantages, but any visual attention model cannot achieve higher quality image region of interest extraction

Images