Motion Deblurring Research Based On Irradiance

Motion blurring is caused by the relative movement of camera and scene during the exposure. Blurring happens at the process of camera exposure after which we get the irradiance image, Then we get the final intensity image via the mapping of camera response function (CRF). But the majority of motion deblurring algorithms do not consider the nonlinear CRF. In addition, the non-linear saturation process leading to saturated pixels appears when obtaining the irradiance image. Image deblurring algorithms without removing saturated pixels are difficult to be robust because the saturated pixels are not blurred as the usual way of the non-saturated pixels. This thesis studies the robust motion deblurring and focuses on these two factors affecting the pixel intensity: CRF and saturation process. The main work is the following:First, the status quo of deburring, the CRF estimation and the saturated pixels are introduced with details. This thesis introduces the mainstream algorithm of non-blind and bind deblurring respectively. The state-of-art CRF estimation method for deburring is also introduced, which is based on exposure ratio, edge accumulation function, etc. We analyze the limitation when applying on motion blurred images. Also, we introduce the handing method to saturated pixels in deblurring.Second, a CRF estimation method is proposed for robust motion deblurring of one or multiple images.In the community there is no motion deblurring method considering the effect of CRF and, correspondingly, existing methods cannot effectively deblur the motion blurred image. We present a new CRF estimation model reflecting the energy accumulation during the motion process during capturing the motion blurred image. The CRF estimation method for one or multiple motion blurred images based on this model is then introduced. Our method simply requires edges non-parallel to the local motion direction, which is more flexible than the previous method that additionally needs the edges be parallel and of high quality. Experimental results show the efficiencies of the proposed CRF estimation model and method for motion blurred image.Third, a block based detection algorithm for saturated pixel sand an improved motion deblurring algorithm based on the combination of CRF estimation and saturated pixel detection algorithm are presented for robust single image motion deblurring. This saturation effect will affect the deblurring results. However, there are few algorithm consider this effect. We propose a block-based algorithm of detecting saturated pixels and applying the detection result to obtain the blur kernel and fulfill deconvolution. The experiment indicates that our method of detecting the saturated pixels improves the efficiency of the motion deblurring. Existing motion deblurring algorithms might be of bad quality without considering the effect of CRF and saturation. A new deblurring method which combines CRF estimation, saturation pixel detection together is put forward. CRF is estimated first to get the blurred irradiance map. Then the saturated pixels are detected and the blurred irradiance map are de-convolved finally to obtain the clear image. This method is more conformed to the formation of blurred digital image than existing methods.