Image local characteristic descriptor

Abstract

The invention discloses an image local characteristic descriptor which is called a Bipond characteristic. The characteristic is not about simple counting of a histogram composed of LBP values of pixel points in a certain area but is about counting of the number of points whose LBP values satisfy a certain attribute I in the area, the attribute being specified by an eight-bit unsigned number (the unsigned number has and only has two bits of 1). The characteristic provided by the invention is easy in calculation and easy to realize, does not require such operation as normalization and the like, is faster than a Hog characteristic in calculation speed, and is higher than the Hog in detection precision.

Description

technical field [0001] The invention belongs to the field of computer vision and image processing, and specifically proposes an image local feature descriptor. Background technique [0002] Image feature design is an important aspect of research in the fields of computer vision and image processing, and combining appropriate features with classifiers is a standard architecture in this field. For example, the combination of Haar features and cascaded Adaboost classifiers makes the face detection technology basically reach the practical level, and the combination of Hog features and SVM classifiers greatly improves the accuracy of pedestrian detection. [0003] The features that are widely used now include Haar features, Surf features, Centrist features, and Hog features. The Haar feature calculates the gray level difference of several adjacent blocks, so it is too simple to achieve the detection accuracy, so it is only suitable for simple applications such as face detection;...

AI Technical Summary

Problems solved by technology

The Haar feature calculates the gray level difference of several adjacent blocks, so it is too simple to achieve the detection accuracy, so it is only suitable for simple applications such as face detection; the Surf feature counts the gray level difference of multiple points in the area, although its The discrimination is better than the Haar feature, but it is still biased and cannot meet the detection requirements; the histogram formed by the LBP value of each point in the Centrist feature statistical area has better performance than the Surf feature; the Hog feature measures the gradient direction of each point in the area. The histogram of the above features has the best discrimination, so it has been widely used in the academic field in recent years, but Hog features need to be normalized and other operations during calculation, which greatly slows down the detection based on Hog ​​features The operating speed of the device is difficult to meet the real-time requirements

Images