The Detection Of Illegal Image Based On Image Automatic Annotation

With the rapid development of Internet and computer technology, resources on the Internet can be acquired easily by people. The easy acquisition of the Internet information has the incomparable advantage which makes the spread of Internet culture fast and wide. At the same time, new life of traditional social culture is also gained so that people’s mental treasure is enriched. However, the proportion of illegal and harmful information has rose which makes very bad influence on people especially the teenagers. Identifying and filtering these illegal information has become an urgent issue. Automatic image annotation is the key in image detection and Image retrieval technology. These labels can imply the semantic content in the image and apply them in large-scale image retrieval. This thesis begins with color space and introduces the basic methods of image processing. The CRF which has many advantages over traditional generative models is introduced in this study. Also the CRF model has the disadvantages such as the model complexity and time consuming progress of model training. The SVM model which has the advantage of low complexity is introduced. This model has a good research significance. Finally the images are detected through automatic annotation.Digital image is saved by pixels which has the disadvantage of redundancy and large data size. That makes the processing of image very difficult. In order to lower the complexity of the algorithm, super-pixels technology which is based on the relationships of the adjacent pixels is used to split the image so that the complexity is reduced. Under the issue of Internet safety and after studying the current situation on technology of image processing, this thesis comes up with a scheme based on super-pixels segmentation and SVM automatic annotation using the study in color space, image preprocessing, face detection, skin color detection. Color space, the transmission of color space, the pretreatment of image, face detection, skin color detection are introduced in detail in this thesis. Meanwhile, the super-pixels segmentation is also elaborated. The experiment results show that this program has the ability to detect the illegal image and also perform well according to the simplification principle.