Real-Time Weighted Median Filtering with the Edge-Aware 4D Bilateral Grid
Weighted median filtering is a fundamental operator in a great variety of image processing and computer graphics applications. This paper presents a novel real-time weighted median filter which smoothes out high-frequency details while preserving major ed
- PDF / 2,873,576 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 95 Downloads / 187 Views
Intelligent Information Systems Institute, Wenzhou University, Wenzhou 325035, China 2 Digital Media and HCI Research Center, Hangzhou Normal University, Hangzhou 311121, China [email protected]
Abstract. Weighted median filtering is a fundamental operator in a great variety of image processing and computer graphics applications. This paper presents a novel real-time weighted median filter which smoothes out high-frequency details while preserving major edges. We define a new 4D bilateral grid by incorporating the 3D bilateral grid with an additional range dimension. The edge-aware weights and the weighted median values are computed in the 4D space. The proposed algorithm is highly parallel, enabling real-time GPU-based edge-aware implementation. Experimental results show that our algorithm can be run efficiently in real-time on modern GPUs. Applications including JPEG artifact removal and image stylization are demonstrated to verify the feasibility of the proposed weighted median filtering algorithm. Keywords: Median filter
1
· Bilateral filter · Bilateral grid · GPU
Introduction
Edge-preserving smoothing filters aim at reducing image noises while preserving meaningful structures. The research on developing new effective edge-preserving smoothing filters have been conducted for decades. The median filter [13] has been proven to be more effective than a linear smoothing filter in preserving edges. Each neighbor pixel contributes to equal weight to the filtered central pixel in the standard median filter. Therefore, it still cannot preserve high-contrast boundaries. The bilateral filter [11] is a widely used edge-preserving filter in image processing and computer graphics community and many improvements have been made in recent years. However, due to the nature of weighted mean of neighboring pixels, noises are also used for weighting the output value of the filtered pixel. On the other hand, the weighted median filter directly selects the neighbor pixel with a weighted median intensity value to substitute the intensity of the filtered pixel. Consequently, weighted median filtering can produce better edge-preserving smoothing results than bilateral filtering in many applications. In recent years, many algorithms have been proposed in order to improve the filtering efficiency. Zhang et al. [17] proposed a few efficient schemes to achieve c Springer International Publishing Switzerland 2016 A. El Rhalibi et al. (Eds.): Edutainment 2016, LNCS 9654, pp. 125–135, 2016. DOI: 10.1007/978-3-319-40259-8 11
126
H. Zhao et al.
a linear computation complexity O(r) with regard to the kernel size r. However, their algorithm is designed for fast CPU implementation but not suitable for GPU parallelization. Ma et al. [6] proposed a constant time weighted median filtering algorithm by elevating the computation in the 3-dimensional space. The constant time guided filter [3] is employed to filter the data in the 3D volume with edge-preservation. Similarly, Yang et al. [15] formulated both the median and bilateral filtering as a unified constant-time cost
Data Loading...
