Time series modeling for an adaptive visibility improvement of the outdoor image sequences in wavelet domain

Under a complex atmosphere, the visibility of images captured by a moving camera needs to be enhanced so as to overcome various atmospheric perturbations. To achieve a stable and robust performance, in this paper, we propose to build a time series based model in wavelet domain by employing both spatial and temporal information of the sequential images. First, we set up a blind evaluation criteria of the image quality (IQ). Second, to each frame of the sequence, we utilize an adaptive computation framework to improve the performance of the wavelet based algorithm, which is controlled by the criteria above. After that, we build two time series models separately, i.e., an autoregressive integrated moving average (ARIMA) model and an autoregressive conditional heteroscedasticity (ARCH) model to fit the parameters change of the complex atmosphere. Finally, we utilize these two models to guide and improve the adjustment of the adaptive calculation. Experiment results demonstrate the effectiveness and robustness of our solution in many outdoor applications.