Summary: | 碩士 === 國立臺灣科技大學 === 資訊工程系 === 102 === In this thesis, an effective median filter- and inpainting-based method for high density impulse noise removal is proposed. We first utilize the histogram distribution to identify the possible corrupted pixels. For each corrupted pixel, we count the number of uncorrupted neighboring pixels, and then a sorting process is applied to schedule the order of noise removal. For the current corrupted pixel, when the number of corrupted neighboring pixels is larger than the specified threshold, an inpainting technique is applied to remove the noise; otherwise, a median filter is applied. Based on four typical test images, experimental results demonstrate that for high-density impulse noise circumstance, the proposed noise removal method outperforms the state-of-the-art method by Hong et al. and some other related methods.
|