LOCALLY ADAPTIVE FILTERING OF IMAGES WITH USING TETROLET TRANSFORM

Abstract

Image filtering is one of the main tasks in image processing. Images are inevitably subject to noise during image formation and subsequent transmission. Thus, it is desirable to remove noise. Image denoising (filtering) improves visual appearance and facilitates subsequent automatic processing (segmentation, classification, detection of edges). A large number of filters has been developed so far. Among them, filters based on orthogonal transforms as well as non-local filters are the most effective. One of the representatives of filters based on orthogonal transforms is the standard sliding window DCT filter. Its effectiveness differs only slightly from the best non-local filters. Non-local filters use search of similar blocks in order to perform collaborative filtering for collected blocks. Due to this, non-local filters require significant computational costs. However, practically all filters run into difficulties in edge/detail preserving. Very often heterogeneous image regions (such as edges, fine details and textures) after denoising seem smeared despite the high noise suppression efficiency. Such a problem is of great importance for segmentation and classification tasks. Because of this, it is expedient to detect such regions and process them without losing useful information. One of techniques able to efficiently preserve edges is the tetrolet transform based filter, nevertheless its noise suppression efficiency is significantly inferior to the DCT filter. In this paper, we propose a locally adaptive filter able to efficiently suppress additive white Gaussian noise and, at the same time, to preserve edges and fine details. The proposed filter is a combination of the DCT filter and tetrolet-based filter, where edge-detail blocks are processed using tetrolet-based filter. In particular, this approach consists of heterogeneity detection and weighting of DCT based and tetrolet transform based filter outputs for the detected areas. Optimization of the locally adaptive filter parameters is carried out. Performance analysis of proposed filter and the DCT filter is done using visual quality metrics. It is demonstrated that the proposed filter provides good edge and fine details preservation capability.

Keywords

• локально-адаптивний фільтр, тетролет-фільтр, дкп фільтр, адитивні перешкоди, детектор неоднорідностей.