Spotting And Classification Of Spontaneous Expression In Video

Compared with traditional expression,spontaneous expression can reveal a person's true feelings and has great potential application in national security,medical diagnosis,etc.The scale of spontaneous expression dataset is relatively small because spontaneous expression is difficult to be induced and collected.At the same time,spontaneous expression has characteristics of shorter duration and weaker intensity,so it is more difficult to recognize and extract features than traditional expression.This dissertation proposes two methods about spotting frame of spontaneous expression in the video and discriminating the categories of spontaneous expression for studying characters of spontaneous expression.(1)Due to the special characteristic of spontaneous expression,this dissertation proposes fine matching algorithm as a preprocessing algorithm for aligning faces in the videos based on ASM algorithm.By this way,the performance of features extracted from faces will improve because those features are less subject to inaccurate positioning and shifting head.In order to make features have higher spotting accuracy and strong robustness,this dissertation proposes an extraction method of local binary pattern spatial-temporal feature with removing redundancy of feature points.It is to extract the local binary pattern feature of the sector region in the spatial axis plane and the local binary pattern feature of the linear feature points without redundant points in the temporal axis plane.The spatial features and temporal features are extracted on the eye region and the mouth region respectively.Then feature values can be obtained by converting features which are made by fusing spatial features and temporal features.Finally,the spontaneous expression frame can be spotted according to comparing the feature value with threshold.The experimental results show that the proposed spotting method can effectively improve the accuracy of spotting frames of spontaneous expression in spontaneous expression video.(2)This dissertation proposes a classification method of deep transfer learning network for identifying the categories of spontaneous expression.Original RGB images and three-dimensional images composed by optical flow feature are trained on different transfer leaning networks for optimization parameters.Then,a new network model,the isomorphic network,is composed by two transfer learning networks of same type that are trained by different samples respectively.Finally,testing results of isomorphic network which is tested by testing samples are compared with the testing results of different single transfer leaning networks.The experimental results show that the proposed method exhibits excellent classification performance of spontaneous expression on different spontaneous expression databases.