An Object Prediction Region Optimization Method Applicable to Target Recognition

Abstract

The invention discloses an object prediction area optimization method suitable for target recognition. The method expands the object prediction area, then performs image segmentation, then performs superpixel bounding box calculation and superpixel saliency evaluation, and finally superpixel-based Sliding window search, and finally the optimized object prediction area is obtained. The present invention can increase the scale of image segmentation by controlling the size of superpixels, thereby reducing the search range of superpixels, so the time complexity of the algorithm is small, which is only related to the number of superpixels in the image; because the pixels in superpixels have Consistency, and the segmentation of local edges is better, so compared with a single pixel, using superpixels as the basic elements of sliding window search can produce better positioning results; the present invention can effectively reduce the search area, thereby accelerating traditional methods based on A sliding window search algorithm for pixels; in addition, by cascading the method of the present invention, the recognition accuracy of the existing target recognition algorithm for the target can be improved.

Description

【Technical field】 [0001] The invention belongs to the field of image processing and computer vision, and in particular relates to an object prediction region optimization method suitable for target recognition. 【Background technique】 [0002] Vision is an important way for human beings to obtain external information, and images are an important carrier of information. With the development of image processing technology, the size and resolution of images are gradually increasing, and the information contained in them is also constantly enriched. Studies have shown that when humans observe an image, their eyes usually move between the objects contained in the image, and they are not interested in other areas such as the background. Most visual technologies, such as pedestrian detection, face recognition, target tracking and target recognition, etc. , which also acts on the above-mentioned region containing the object. Therefore, how to quickly and effectively locate the posi...

AI Technical Summary

Problems solved by technology

However, the method based on sliding window search mainly faces two problems in practical application: 1. Since the entire image needs to be searched and the classifier is applied in all possible positions, the time complexity of the algorithm is relatively high; 2. How to effectively train Classifiers with position discriminative features still need further research

For the first problem, most of the existing improvements use heuristic algorithms to speed up the search process, but it will increase the probability of false detection of object areas

Images