Research Of Digital Clock Recognition In Video Based On Machine Learning

The digital clock is an important application problem, and has important application value in the field of intelligent monitoring and entertainment. The existing algorithms in the literature can be used to read the digital clock in the video, but there are two problems in this algorithm. The first is that these algorithms are mainly based on heuristic methods, but do not use powerful machine learning methods. Second,these algorithms have a number of parameters need to manually adjust.To solve these two problems, this paper proposes the use of machine learning methods to complete the clock positioning and clock digital recognition. This paper is the first study on the identification method of machine learning. The idea is as follows:different from traditional methods, in the localization phase, according to the periodicity of the second pixel gray value changes quickly locate the position of the second digital pixel, and first introduced machine learning theory to solve this problem,more intelligent. Secondly, considering the digital clock usually has a stable background, and different digital clock foreground and background colors, quickly locate the clock and sub regions, first using machine learning to distinguish colors, and then using machine learning for image segmentation, but because of the use of the color information, higher efficiency.In this paper, we use the convolution neural network to recognize the single digit,and get a higher efficiency than the CRF, and reduce the dependence on the font.Because the number of seconds has a series of characteristics, this study also made a series of identification, which is used to correct the error of single digit recognition.Finally, the research on how to draw lessons from the previous heuristic function of the results of the experience, how to use the machine learning method for the network structure design for specific problems, and Simulation of heuristic function, a discussion.