Face Recognition Method for Handheld Devices

Abstract

This patent proposes a face recognition method that combines adjacent LBP features and basic LBP feature extraction to train a cascade classifier for handheld devices. The method of the present invention includes: first, performing feature extraction and classifier training on a training sample database based on LBP and combined adjacent LBP features, to obtain a cascaded classifier including N classifiers, wherein the number of features for training each classifier is graded step by step Increase, the first N-1 classifiers use LBP for feature extraction, and the Nth-level classifier uses combined adjacent LBP features for feature extraction; then extract the basic LBP features of the image to be recognized as the input of the cascade classifier for step-by-step classification It is determined that when the Nth-level classifier is determined, it needs to be converted into a combined adjacent LBP feature, and when the thresholds of each classifier are met, it is determined to be a face image. The invention is used for face detection on hand-held devices, and has the characteristics of fast calculation speed and high recognition rate.

Description

technical field [0001] The invention belongs to image processing technology, in particular to a face detection technology used for handheld devices. Background technique [0002] With the continuous development of the market, face detection on handheld devices (such as mobile phones, PDAs, etc.) has received more and more attention. Although many algorithms for person detection exist, only a few of them address the problems encountered in face detection on handheld devices. Face detection on handheld devices is different from normal face detection: first, most of the face images are frontal images, with few movement changes; second, most of the images are close-up images with high resolution. Because the images collected by the handheld device have the above characteristics, only a small number of video frames can be processed to obtain the ideal detection effect. However, in the process of image acquisition, there are also the following problems: (1) The mobility of the d...

AI Technical Summary

Problems solved by technology

However, in the process of image acquisition, there are also the following problems: (1) The mobility of the device leads to drastic changes in lighting conditions, and the performance of many existing face detection algorithms deteriorates as a result.

(2) Compared with other high-performance devices, such as PCs and workstations, handheld devices have limited computing power. If the algorithm is too complicated, even if the recognition rate is ideal, the computing speed is difficult to satisfy

Hadid et al. have successfully applied the LBP histogram feature to face detection, but only using the LBP feature to train the classifier for face detection has the defect of fast detection speed but low accuracy.

In order to achieve the ideal detection accuracy, dozens of levels of cascading enhancements are required, which brings great inconvenience to the performance allocation of classifiers, the selection of weak classifiers and parameter adjustment, and the training process of classifiers also becomes extremely complicated.

Images