Research And Implementation Of Image Quality Test System For Multiple Video Streams

In order to ensure the safety and image quality of broadcast, TV station need to test the pictures of program at technology and content level. The traditional manual monitoring method has the problem of false rejected, false alarm, low efficiency and so on. The related products developed by some enterprises maybe meet the requirements on some levels, but there are still existed the following problems:1. They can only detect some simple items such as black-frame, green-frame and color-bars, they can’t detect other complex items like mosaic.2. Their methods only consider single image but without considering the continuity of the video contents, and they has a fewer accuracy.3. As the amount of computation in image processing is huge, it is detecting slowly by the limited CPU and it can only detect fewer streams at the same time.Aiming at the above problems and combined with a higher requirement for image quality by TV station, This paper has carried on in-depth discussion and research based on the existed methods and systems, the mainly work are listed follow:1. Studying on the improvement of existed detect methods and create the detect methods for complex items such as clip-frame, snowflake, shake and mosaic. For simple items, this paper improves the accuracy by using the existed methods combined with considering continuous of the video contents. Using similar scenes and image upheaval for clip-frame detection; Using counting of connected region of absolute difference image for snow detection; Using tracking of feature-point and analysis of motion trend for shake detection; Using cherry-picking of image region and search of equidistant color-block for mosaic detection.2. Researching on hardware acceleration for image quality detection based on GPU. GPU has higher concurrent computation ability than CPU, dividing image into small blocks and send to GPU for parallel processing when process don’t need consider block’s correlation, this method can improve the detection speed greatly.3. Designing and realizing the system software. The image quality detection realized by some frameworks like OpenCV, FFmpeg and CUDA. System software also includes some functions like real-time recording for management personnel can see it later, image rendering with custom rendering model for field monitor and real-time pushing of image for remote monitoring.The test result shows that the method and the system used in this paper can solve the problems of the present system and finally achieve the design target of image quality detect system which needs high accuracy, low cost, stable, safe and extensible.