Research Of Qucik Algorithm To Extract ROI Form Video And Its Application

The parts in a picture which are able to attract people's attention and consist of key information are called the region of interest. With extracting ROI of image, we can reduce calculation of computer and improve the efficiency. With applying in image compression and coding, object recognition, image retrieval and so on, ROI extraction is worthy to research.Compared with image, video is incorporated with time information, it is more complex. Applying traditional visual attention models on video will lose move information. Besides, the data of video is usually very huge, the algorithm will lose its significant of application if the processing speed is very slow. To application, the algorithm also needs high precision. An algorithm to extract ROI from video quickly and accurately based on the contents of existing models is proposed in the paper. The main contents of this paper are as followed:(1)The status of ROI extracting methods is introduced. The method based on visual attention models is the research focus. So, three visual attention models are stated in details. The fastest model is chosen as the base of our algorithm.(2)With Designing eye movement experiment for videos which are needed and selecting eye movement data, the data of fixation is obtained. Then with Gaussian distribution simulation, the eye movement ROI is extracted. It is used as the real ROI of image. The ROI is applied to evaluate other methods.(3) The SR model is improved from both spatial and temporal cues. In terms of space, with the analysis of saliency information in frequency domain, the result which is obtained from the convolution integration between Gaussian kernel and amplitude spectrum is added to the model. The original model only consider the intensity feature, color features are included to improve the accuracy of the model. In addition, hyper-complex Fourier transformation is used to reduce computation time. After that, with the analysis of image similarity, reduplicate calculation was avoided. In terms of time, frame difference is used to extract motion feature. In the end, an objective index is used to evaluate the efficiency and accuracy of our algorithm.(4)The ROI extracted by our algorithm is applied to H.264 video coding standard. With the analysis by PSNR between two videos which are compressed to the same bitrates, we find that using ROI in video compression and coding can improve the quality of compressed video.A new algorithm based on SR model to extract ROI from video is proposed in this paper. Experiment shows that the algorithm is more accurate than three visual attention models which are introduced in the paper. Also, when the algorithm is applied to a 10-seconds video, it only costs 15.4S which is close to the time SR model needs. The resolution of the video is 1920 x 1080 and the frame rate is 30FPS. It means the algorithm is quick and accurate.