A survey on visual adult image recognition

ORB feature based web pornographic image recognition

“…But because of the insufficient of passivity, hysteresis, and high error rate of recognition, those methods based on keywords or IP/URL have not been widely adopted. At present, those methods based on the content of images have become the mainstream methods in the research area of web pornographic image recognition [7,9,11] , and can be subdivided into three categories: the first category is to analyze or understand the human body movements in the image with reference to the human body language, and they can determine whether it is pornographic image or not according to pre-determined rules or models [10] . But these kinds of methods do not always understand the image successfully and cannot obtain a satisfactory recognition result due to the complex and diversity of the pornographic images.…”

“…But these kinds of methods do not always understand the image successfully and cannot obtain a satisfactory recognition result due to the complex and diversity of the pornographic images. The second category is those based on Content Based Image Retrieval (CBIR) [11] which usually constructs a predefined image database containing vast amount of labeled pornographic and non-pornographic images. The image to be recognized is used as the query image, and the retrieval images from the database are analyzed.…”

Skin information combined saliency detection for classifying objectionable images

“…Therefore, many studies to block improper image by visual content has been studied in the last years. The current approaches can be divided into three main groups based on their underlying ideas [1]. The first group of approaches classifies the objectionable image based on color information.…”

“…The second group of approaches relies on shape information. These approaches use the shape features such as contour-based feature, moments and MPEG-7 features [1]. The last group of approaches uses the local feature descriptors to recognize adult image.…”

“…The method proposed by Achanta et al [6] used for visual attention model. The saliency value at the give pixel S ss (x,y) is obtained as : (1) where I μ (x,y) is the average CIELAB vector of the symmetric surround region and I f (x,y) is the corresponding CIELAB image pixel vector in the Gaussian filtered version of the original image. The skin probability map is generated using the probability of skin and non-skin color.…”