Adaptive mosaic image representation for image processing

TL;DR

This paper introduces an adaptive mosaic image representation method that uses a multigrid algorithm to efficiently approximate image regions of varying complexity, aiding in tasks like edge detection and image compression.

Contribution

It presents a novel multigrid-based mosaic image representation technique that adaptively segments images based on local transition frequency for improved processing.

Findings

Effective for contour and edge detection

Useful for lossy image and video compression

Adapts to image non-homogeneity levels

Abstract

Method for a mosaic image representation (MIR) is proposed for a selective treatment of image fragments of different transition frequency. MIR method is based on piecewise-constant image approximation on a non-uniform orthogonal grid constructed by the following recurrent multigrid algorithm. A sequence of nested uniform grids is built, such that each cell of a current grid is subdivided into four smaller cells for the next grid designing. In each grid the cells are selected, where the color intensity function can be approximated by its average value with a given precision (thereafter 'good' cells). After replacing colors of good cells by their approximating constants the reconstructed image looks like a mosaic composed of one-colored cells. Multigrid algorithm results in the stratification of the image space into regions of different transition frequency. Sizes of these regions depend on the few tuning precision parameters that characterizes adaptability of the method to the image fragments of different non-homogeneity degree. The method is found efficient for prominent contour (skeleton) extraction, edge detection as well as for the Lossy Compression of single images and video sequence of images.