Intensity Classification of the Image Quality Metric Series Using AdaBoost and Co-training Method

In this paper, the AdaBoost and the co-training method are utilized to analyze the intensity change of the Image Quality Metric Series (IQMS) of the outdoor moving camera. Since the image quality will affect the subsequent Computer Vision (CV) algorithms seriously we use the time series analysis technique to research its change laws: 1) we calculate the image quality metrics, the contrast level, the blur level and the noise level, to represent the definition of an image. 2) We cumulate enough computation results of these metrics above in the time axis. 3) We employ the time series indexing technique to mine the change pattern of these series to get an indexing description of them. 4) We use AdaBoost classifier and co-training method to distinguish the change intensity of these series. After the process steps above, we can segment the image sequence into different parts according to their change pattern of image quality. Then these parts can be processed by some different but proper CV algorithms for outdoor application. Many experiment results have shown the validity of our proposed method.