Internet porn image detection method based on deep convolution nerve network

Abstract

The invention relates to an Internet porn image detection method based on a deep convolution nerve network. The method comprises the following steps of acquiring a porn image and a normal image through a manual calibration method and carrying out pretreatment and enhancement on the images so as to acquire an effective square training image; sending the acquired effective image into a deep convolution nerve network so as to train the network; verifying a network model on a verification set, adjusting a training set according to a result and continuously training the deep convolution nerve network; repeating the last step till that detection accuracy on the verification set reach an expected object or a network loss function begins convergent; testing the trained network on the training set. The method in the invention has the following advantages that the porn image detection method based on the deep convolution nerve network is provided and the method can be used to rapidly detect almost all the types of porn images through a single model; in an actual test, detection accuracy in the invention reaches above 98.6%.

Description

technical field [0001] The invention belongs to the fields of computer vision and image recognition, and relates to a network pornographic image detection method based on a deep convolutional neural network. Background technique [0002] Existing specific methods for accurately and quickly detecting pornographic images from massive network images are as follows: [0003] The first pornographic image detection method based on skin color statistics [0004] Solution: Use the skin color detection model (based on different color space color features or texture features, and use Gaussian mixture model to judge whether a pixel is skin) to detect the skin color area in the image, and judge according to the ratio of the skin color area to the total image area Whether the image is pornographic. [0005] Disadvantages: (1) Many images are not pornographic images even if there are more skin areas (such as images containing characters in bikinis, such as figure 1 (a)); [0006] (2) ...

AI Technical Summary

Problems solved by technology

[0005] Disadvantages: (1) Many images are not pornographic images even if there are more skin areas (such as images containing characters in bikinis, such as figure 1 (a));

[0006] (2) The proportion of the skin color area in some pornographic images is not high in the whole image (some pornographic images only expose pornographic parts);

[0010] Disadvantages: (1) Which type of image to choose to build the image database to be retrieved has a great impact on the final effect (both pornographic images and normal images have a large number of styles, it is impossible to select limited images to represent all types, and there will often be some A pornographic image cannot retrieve a very similar image in the image database);

[0011] (2) It is very difficult to select which features to compare the similarity between two images (weak features will bring a lot of misjudgments, and strong features will cause similar pictures to be retrieved)

[0014] Disadvantages: (1) The skin color detection model is also used, which has the same defects as the existing method 1;

[0015] (2) Which low-level visual features to use and how to choose a classifier are based on experience, and it is difficult to guarantee the best detection effect

[0018] Disadvantages: (1) It is difficult to use which features to construct a visual word bag;

[0019] (2) There is no corresponding evaluation standard for the size of the bag of visual words;

[0020] (3) The proportion of the main part of some images is too small, and a large number of meaningless feature points will appear through the extraction of feature points, which will affect the classification effect; figure 1 As shown in (d), if this is a pornographic image, the extracted feature points will contain a lot of disturbing background information, which will affect the judgment of the classifier

[0023] Disadvantages: The number of training samples is small, and it is easy to produce overfitting, which will lead to high accuracy in the training set and low accuracy in the test set

Images