Focusing on the issue of target motion feature extraction of video sequences in complex scene, and referring to the motion perception of biological vision system to the moving video targets, the traditional primary Visual cortex (V1) cell model of visual cortex was improved and a novel method of random-dot motion feature extraction based on the mechanism of biological visual cortex was proposed. Firstly, the spatial-temporal filter and half-squaring operation combined with normalization were adopted to simulate the linearity and nonlinearity of neuron's receptive field. Then, a universal V1 cell model was obtained by adding a directional selectivity adjustable parameter to the output weight, which solved the problem of the single direction selectivity and the disability to respond correctly to multi-direction motion in the traditional model. The simulation results show that the analog outputs of proposed model are almost consistent with the experimental data of biology, which indicates that the proposed model can simulate the V1 neurons of different direction selectivity and extract motion features well from random-dot video sequences with complex motion morphs. The proposed method can provide new idea for processing feature information of optical flow, extract motion feature of video sequence and track its object effectively.

The deformation such as translation, rotation and random scaling of local images in image recognition tasks is a complicated problem. An algorithm based on pre-training convolutional filters and Multi-Scale block Convolutional Neural Network (MS-CNN) was proposed to solve these problems. Firstly, the training dataset without labels was used to train a sparse autoencoder and get a collection of convolutional filters with characteristics in accord with the dataset and good initial values. To enhance the robustness and reduce the impact of the pooling layer for the feature extraction, a new Convolutional Neural Network (CNN) structure with multiple channels was proposed. The multi-scale block operation was applied to input image to form several channels, and each channel was convolved with corresponding size of filter. Then the convolutional layer, a local contrast normalization layer and a pooling layer were set to obtain invariability. The feature maps were put in the full connected layer and final features were exported for target recognition. The recognition rates of STL-10 database and remote sensing airplane images were both improved compared to traditional CNN. The experimental results show that the proposed method has robust performance when dealing with deformations such as translation, rotation and scaling.