High performance non-linear filter for impulse noise removal and edge detection in images

Impulse noise is one of the prevalent noises that undermines images. It produces noise of a brief duration, and is often of severe amplitude during a brief interval of time. Whether your images are acquired with a mobile phone, high-end digital camera or scanned, they are susceptible to impulse noise. Removing impulse noise is a prerequisite process for many higher level tasks in image processing such as image compression, image coding, and object recognition. However, noise removal always corresponds to a loss in some image details. Hence, the reduction of noise while preserving the fine details of the image has attracted the attention of many authors. Therefore, a multi-phase homogenization (MPH) technique based on the similarity criterion is proposed to suppress and restore random-valued impulse noise in color and gray-level images. In particular, a four-phase method is introduced and discussed. An image pixel is considered to be an original if it satisfies a threshold criterion at each phase. The criteria at different phases are consistent in the sense of being logically nested. If a pixel does not satisfy the criterion at a given phase, it is deemed noisy and replaced by using a weighting procedure, and thus homogenized. Compared with state of the art methods, the proposed approach illustrates comparable results at low noise rates, but it is clearly superior to other methods at high noise rates. MPH technique also maintains a low computational complexity and effectively preserves image details. The same principle of the MPH method is extended to be used for edge detection in blurred, noisy and noise-free images. In that sense the pixels that have minimum numbers of similar pixels within the local window and in a predefined intensity range are labeled as edge points. Thus, better results than many previously known methods are shown by the proposed edge detector.