Qing Wang

A random-valued impulse noise removal algorithm with local deviation index and edge-preserving regularization

In this paper, we present a two-phase random-valued impulse noise removal algorithm based on local deviation index (LDI) and edge-preserving regularization. In the first phase, we define an image statistic LDI. Then with image pixels’ LDI values, the outlier candidates are identified. In the second phase, the image is denoised by an edge-preserving regularization method. Extensive experimental results indicate that our method performances better than many existing filters do for its robust image restoration and accurate noise detection.

An adaptive median filter using local texture information in images

After evaluating performances of the directional median (DM) filter and the adaptive switching median (ASWM) filter, we propose an adaptive median filter for restoring images by using local texture information in images. It contains two steps: 1) identifying noise pixels; 2) estimating the values of the noise pixels. Firstly, a double-layer window is adopted to improve the adaptive switching median filter, and the inner layer is used to detect noise pixels while the outer layer is used to obtain local texture information. According to the texture features, the detected noisy pixels are then restored by the center weighted median filter. Finally, experimental tests are done on evaluating the algorithms time consumption, noise detecting rate, and restoration quality. The test results show our algorithm has satisfying performance on restoring high-corrupted images and is with lower time consumption compared to the existing approaches.

Edge-Preserving Regularized Filter with Spatial Local Outlier Measure and Q-Estimate

This paper presents an efficient random-valued impulse noise removal algorithm. The filtering process contains two phases: a detection phase followed by a filtering phase. In the detection phase, the proposed method uses the novel image statistics, the spatial local outlier measure (SLOM) and the Q-estimate, to identify impulses in a corrupted image. When the noise pixels are identified, their values are restored by an edge-preserving regularized method in the filtering phase. Extensive experimental results show that our filter provides a significant improvement over many other existing techniques.

An Efficient Detail-Preserving Impulse Noise Filter with ROLDM

This paper presents an efficient detail-preserving random-valued impulse noise filter. The proposed method introduced an image statistic, the Rank-Ordered Logarithmic Difference from median (ROLDM for short), to ditinguish image details from impulses. This reduces the probability of detecting image details as impulses. In order to search for suitable thresholds at noise detecting phase, we present the Q-estimate of variance in noisy image. According to image variance, we define a threshold for each pixel. This makes more impulses to be identified. Experimental results show that our filter provides a significant improvement over many other exsiting techniques.

Detection of Man-Made Structures in Natural Images Based on the Improved Discriminative Random Fields

This paper presents a new Discriminative Random Fields (DRFs) framework. Based on the DRFs framework proposed by Kumar and Hebert, the following improvements have been conducted. Firstly, the interaction potential and the associated potential model are simplified. Secondly, we reduce the dimension of the multi-scale features, re-define dimension of the single-scale feature, and increase the color feature of man-made structure. Thirdly, the quasi-Newton method with linear search and gradient descent method are adopted to solve parameters, which get a simple model and achieve good performance. Finally, the partition function of the DRF is eliminated by using Pseudo-likelihood method for parameter learning. The simulation results show that the proposed methods false positive rate is lower than the method from Kumar and Hebert, while the correct rate and detection ratearehigher than their experimental effects after these improvements.

A Two-Step Random-Valued Impulse Noise Removal Algorithm

A two-step random-valued impulse noise removal algorithm is proposed. In the first step, we utilize local image statistics and human visual perception to restore the noise pixels in the flat region. In the second step, we aggregate local useful image information and introduce the rank-ordered absolute difference statistics and edge-preserving regularization to process the impulse pixels near the edges for further. Finally, experimental results show that our approach provides better performance in term of PSNR and MAE than many existing random-valued impulse noise filtering techniques.

An Efficient Pre-Processing Algorithm for Removing Uniform Noise Based on Cluster Method

An efficient pre-processing algorithm for removing uniform noise is proposed. Local image statistic information and human visual perception are used to classify the pixels in the filter window. According to the elements number of each cluster, all pixels are divided to noise-free clusters or fuzzy clusters. Through cluster method, almost all noise pixels are identified and then restored. Finally, we choose some commonly used filters to test our algorithm. The experimental results tell that our approach can enhance those filters’ capability of suppressing impulse noise effectively. Due to the proposed algorithm can decrease the noise density effectually and keep image details, it can be introduced into many existing uniform noise filtering techniques.