Method for recognizing pornography images on mobile Internet based on multi-mode combinational strategy
A mobile Internet and image recognition technology, applied in character and pattern recognition, instruments, computer components, etc., can solve the problem of high misjudgment rate of normal images, achieve high recognition rate, reduce misjudgment rate, and improve the effect of recognition rate
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0096] Embodiment one: see figure 2 , the mobile Internet pornographic image recognition method based on the multi-mode combination strategy of the present invention first uses the pornographic image recognition algorithm based on wavelet texture correction of skin color to perform the first rough filtering on the image to identify suspected pornographic images, and on this basis first adopts Misjudged image matching technology collects images that are often misjudged, and forms a million-level misjudged image library, then extracts the global features of the image, and uses E2LSH technology to index the images in the image library; The suspected pornographic images are quickly and accurately matched using E2LSH based on the million-level misjudgment image library. After matching, if they are in the image library, they are considered normal images. If they are not in the misjudgment image library, they are considered suspected pornographic images. ;Finally, for the suspected ...
Embodiment 2
[0097] Embodiment two: see figure 2 , image 3 , this embodiment is based on the mobile Internet pornographic image recognition method based on the multi-mode combination strategy. The pornographic image recognition algorithm based on wavelet texture correction of skin color is based on skin color detection. Wavelet texture analysis is added to remove the skin-like points caused by the skin-like background, and finally use The skin-tone area ratio performs a first coarse filter on the image. This technology can maintain a high recall rate of pornographic images, and at the same time meet the requirements of the front-end machine to process more than 200 images per second. The specific process is as follows:
[0098] 1) Transform RGB color space to HSV color space;
[0099] 2) First quantify the HSV color space and divide it into L color subspaces, then determine the distribution of skin color in these L subspaces through statistical analysis, and cluster to obtain the dist...
Embodiment 3
[0111] Embodiment three: see figure 2 . The mobile Internet pornographic image recognition method based on the multi-mode combination strategy in this embodiment is different from Embodiment 1 or Embodiment 2 in that: firstly, images that are often misjudged are collected by using misjudgment image matching technology, and formed into a million-level A library of misclassified images at scale. Then extract the global features of the image, and use E2LSH technology to index the images in the image library. For the image result of the first coarse filtering, E2LSH can be used for fast and accurate matching, and if it is in the image library after matching, it is considered to be a normal image. In this way, the possibility of the same image being misjudged multiple times is avoided, and the misjudgment rate is greatly reduced.
[0112] Locality Sensitive Hashing (LSH) is similar to the nearest neighbor search algorithm. Its basic principle is: for a point in the space, it is...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 