A cascade classifier training method and device

Abstract

Embodiments of the present invention provide a cascaded classifier training method and device for reducing the false detection rate of the cascaded classifier. The method includes: sequentially selecting samples for each level of strong classifier for training, and for any level of strong classifier after the first level of strong classifier, when performing negative sample training, judging whether the size of the selected negative sample is smaller than the pre-set Set the size, if yes, scale the selected negative sample to a picture of the detection window size of the cascaded classifier; otherwise, intercept the picture of the detection window size from the selected negative sample; input the picture obtained by processing the negative sample into the A strong classifier that has been trained before the first-level strong classifier detects, and selects a picture that has been misdetected; and performs negative sample training on the strong classifier of any level according to the picture that has been wrongly detected.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a cascade classifier training method and device. Background technique [0002] Currently, cascaded classifiers are mainly used for image content recognition. The cascade classifier is an iterative algorithm whose core idea is to train different classifiers (weak classifiers) for the same training set, and then combine these weak classifiers to form a stronger classifier, namely a strong classifier. ; Train multiple strong classifiers and combine them to form a cascade classifier. Only images whose features meet the requirements of each weak classifier in the cascade classifier are considered to contain the target content. [0003] Work using cascade classifiers consists of two stages: training and detection. The training process is divided into three steps: first, a certain feature of the training sample needs to be extracted, such as local binary pattern (lbp) feature, oriente...

AI Technical Summary

Problems solved by technology

[0006] The embodiment of the present invention provides a cascade classifier training method and device to solve the problem of high false detection probability of the cascade classifier trained by the current method

Images