Content Based Nonlinear Video Scaling Technique

In recent years, to enhance the display quality of LED panel has become a research hotspot in the industry. The screen size and aspect ratio of LED panel is free to choose by users. It will be distortion, if the screen does not meet the standard ratios in a full-screen play. In this case, how to improve the display quality has become urgent to be solved. The methods in this paper can solve this problem to some extent.First, the author finds the visual focus region of image. It is proposed an improved adaptive Canny algorithm. The improved algorithm, not only do not need to manually input parameters, but also can achieve good results when dealing with noisy images. Then, it has suppressed the stray edges from edge image. To do this, it can significantly reduce the pressure of the following image segmentation.Non-linear scaling can be seen as a proportional scaling and non-proportional tensile combination. And the tensile direction is always along with one-dimensional space. Based on this characteristic, the author has proposed a simple and effective algorithm for image segmentation. This method is edge points projection based, and it can rapidly segment the image as backgrounds and foregrounds. Experiments have proved that this method is not only effective but also very fast.The amplification of background region would be greater than the foreground area. In order to avoid the phenomenon of visual snag when the stretch ratio is too large, the author has proposed a background scaling method which is based on the combination of background virtualization and interpolation amplification. Experiments have proved that it can achieve good results in some images.Finally, the author has used a parallel GPU acceleration technology to meet the real-time requirements, and used DirectShow framework to process the video frames.