Video Key Frame Extraction And Facial Expression Recognition

With the rapid development of internet technology and the embedded devices, such asset-top boxes, network players and so on, a variety of text messaging and streaming mediahave exponentially grown in our daily life. Lots of erotic fictions, images, videos and otherharmful adolescent health information on the internet enable adolescent to form the falsevalues, and then resulting in the adverse influences to the construction of spiritual civilization.In order to reduce the adverse effects of the video on the healthy growth to young people, it isnecessary to study the characteristics and the detection technology of adverse video. And theobjectionable video detection and its characteristics are firstly processed to scenesegmentation, shot segmentation and key frame extraction. In this procedure, we use the keyframe sequence as the main content of the video and then convert the study of poor video asthe research key frame content, which could greatly reduce the storage amount of dataprocessing. Secondly, the key frames can be carried out from the facial expressions,movements, and so on. Based on these properties, in our paper, the expression assisting thepoor video detection is studied.The research group has completed the researches on the poor video recognition, sceneclassification, shot segmentation, face detection, and the initial detection of bad video.Preliminary detection of objectionable video has completed through energy curve of thespecific video, and then build feature library, in which characteristic database is to saveenergy curve and measure video of curve matching. Further, the viewers’ expression can assistthe detection of a particular video. Firstly, the audience related video is recorded and torecognize viewers’ expression and synthesis of initial bad video detection, then determine theperformance of the video. Besides, we can also, warn against inappropriate video based onthe age of the audience. In this paper, based on the above works out research in two aspectsare carried: video key frame extraction and recognition of facial expressions in key frame.In key frame extraction, the key frame extraction method based on visual attention modeland complementary integration is proposed. Firstly, because the single feature is not comprehensive and different features have good complementary, saliency can express somesemantic information, so the brightness of space, local binary pattern (Local Binary Patterns,LBP) space significantly and motion significantly degrees are respectively used to extract keyframes. Secondly, key frames obtained may redundancy, and redundant key frames will not beable to provide information and will result in a waste of storage space. Comparing thecorrelation coefficient thekey frame could be extracted. Besides, through the removal of thecorrelation coefficient larger key frame, the redundancy of key frames is solved to a certaindegree. Experimental results show the method to extract key frames in line with human visualcharacteristics and can express a more comprehensive video content.In the aspect of face recognition, face recognition is proposed based on the improvedsecond-order partial derivative pattern (LDP). Firstly, based on the AdaBoost algorithm, theface in key frames is detected. Due to the LBP is no first-order partial directional modeLDP,however, is able to describe some direction information. And in our paper, the improved LDPfeature extraction is proposed to improve the LDP algorithm to different changes in directionthat can been coded. Finally, since K-nearest neighbor algorithm for the unknown andnon-normal distribution of data can achieve better classification results, the concept is clear,easy to implement. Thus, we use the K-nearest neighbor method to classify the expression.