Training method and device of classifier, and method apparatus for recognising sensitization picture

Abstract

The invention provides a training method and a device for a sensitive picture classifier; the regional shape characteristic of the training picture set is extracted; the distribution characteristics of the regional shape characteristic in the positive example sample set, the first negative example sample set and the second negative example sample set are measured; the divisibility of the regional shape characteristic is determined according to the distribution characteristics; the regional shape characteristic with the divisibility relative to the first negative example sample set is marked as the first characteristic group; the regional shape characteristic with the divisibility relative to the second negative example sample set is marked as the second characteristic group; the first classifier is obtained through the characteristic training of the first characteristic group; the second classifier is obtained through the characteristic training of the second characteristic group; the invention also provides the method and the device which adopt the sensitive picture classifier to recognize the sensitive pictures; the invention improves the accuracy of the sensitive picture recognition.

Description

technical field [0001] The invention relates to the field of image recognition, in particular to a classifier training method and device, and a method and device for identifying sensitive pictures. Background technique [0002] With the increasing amount of information on the Internet, there are also more and more bad information. Sensitive images such as pornographic images in bad information pollute the social atmosphere and endanger the physical and mental health of young people. Identifying and blocking such sensitive images is a key task for purifying Internet content. [0003] Considering that the vast majority of sensitive images have large areas of exposed human skin, and detecting exposed human skin is relatively easier than directly detecting sensitive images, human exposed skin detection is an effective heuristic method for sensitive image recognition. Through human naked skin detection, a class of sensitive images with a high degree of suspicion can be detected,...

AI Technical Summary

Problems solved by technology

[0008]Because in the prior art, all kinds of negative example pictures are combined to form a negative example sample set, the distribution of shape features of some areas of the negative example pictures is further dispersed, increasing The degree of feature overlap between the positive and negative images, for example, many overlapping features of the portrait image and the sensitive image in the negative image are forcibly marked as different labels, resulting in overfitting of the trained classifier, and the occurrence of classification problems. Distortion, the false detection rate of portrait images and the missed detection rate of sensitive images will increase, and the classification results of scene images will also be affected by unpredictable effects

Therefore, the classifiers trained in the prior art have the problems of high missed detection rate and high false detection rate.

Images