A video saliency detection method based on 3D convolution neural network is proposed

The invention relates to a saliency detection method for a video image, characterized in that: firstly, a 2D deep convolution neural network is established by 2D convolution, a video frame is inputtedto obtain the semantic features of the moving object, then a 3D depth convolution neural network is established by 3D convolution, spatio-temporal salience features are obtained by inputting three consecutive frames of video frames, and then the moving object semantic features and spatio-temporal salience information are connected and inputted into a 3D deconvolution network to learn and mix spatio-temporal salience features, and finally a salience map is obtained through the 3D deconvolution network. So we get a saliency map of the whole image, and the greater the saliency, the more salientthe pixel is, that is, the more eye-catching it is. The experimental results show that the video image saliency detection model has excellent detection performance. 本发明涉及种对于视频图像的显著性检测方法，其特征在于：首先利用2D卷积建立个2D深度卷积神经网络，输入帧视频帧获得运动目标