Study On Impulse Noise Removal Of Digital Images Based On Complementary Directions

The image is an entity that directly or indirectly acts on human visual system to produce visual perception and can be divided into two categories, the analog image and the digital image. With the development of computer and digital technologies, digital image becomes a way to reproduce natural things, while digital image processing has been widely studied in recent years.Digital image is often corrupted by different kinds of noise during acquisition, transmission and storage. Impulse noise have greatest effects on image that even a small amount of noise will damage the image seriously, so it is necessary to remove noise before subsequent processing, such as image segmentation, object recognition. Research results indicate that compared with linear methods, nonlinear techniques can provide more satisfactory results. The median filter was once the most typical nonlinear method due to its good denoising power and computational efficiency.Two common types of impulse noise are the salt and pepper noise whose noisy pixels can take only the fixed-values and the random-valued impulse noise whose noisy pixels can take any random values in the dynamic range. In this paper, we firstly propose an improved median filter based on complementary directions to removal salt and pepper noise. We adopt decision-based method for noise detection and choose a pair of complementary directions for noise removal which utilizes unsymmetric trimmed median filter. The final restored image is obtained by summarizing two pre-processed images. This algorithm makes full use of information inside window. Simulation results illustrate that the proposed algorithm has good denoising ability in terms of the qualitative and quantitative measures and outperforms the existing methods especially at high noise ratio. Moreover, we propose an algorithm to remove random-valued impulse noise utilizing boxplot for noise detection. We first introduce boxplot visualization of outlier from data mining to noise detection in which image noise is regarded as outlier, and then combine with complementary direction method to filter the noise. Simulation results verify the effectiveness of the proposed algorithm and both the subjective and objective evaluations indicate that, compared with the traditional classical algorithms, the algorithm is robust to random-valued noise removal as well as preserves the image structural information effectively.