Method for establishing human-face detector

Abstract

The invention provides a method for establishing a human face detector, the key of which is to use the non-human face image detected incorrectly in the training process of the current layer strong classifier as the non-human face image for training the next layer strong classifier. When building a face detector, the classifier is first trained according to the current layer of non-face image collection, the current layer strong classifier is established, and the non-face images detected by errors in the training process are recorded; then the recorded non-face images are used as Part of the samples in the next layer of non-face image collection, and the non-face image is scanned by the current waterfall to construct the remaining samples in the next layer of non-face image collection; the process of training a strong classifier and constructing a non-face image collection is repeated , until the establishment of the face detector is completed. By applying the solution of the present invention, the time for constructing a non-face image set can be reduced, thereby increasing the speed of establishing a face detector.

Description

technical field [0001] The invention relates to image recognition technology, in particular to a method for establishing a human face detector. Background technique [0002] Face detection is a kind of image recognition technology, and it is one of the important technologies to realize machine intelligence. Accurate positioning. At present, the application of face detection is very extensive, and can be applied to technical fields such as face cartoonization and video face changing show. [0003] The construction of classifiers is the core of face detection technology, and each classifier has one input and two outputs. When an image is input into the classifier, the classifier detects the image and judges whether the image is a face image, and if so, outputs the information that the image is a face image; otherwise, outputs the information that the image is a non-face image information. [0004] Classifiers are generally divided into weak classifiers and strong classifie...

AI Technical Summary

Problems solved by technology

[0014] It can be seen that in the prior art, there is no method for quickly establishing a face detector

Images