A variational textured image decomposition model and its efficient MPI implementation

Image decomposition consists of splitting an image into two or more components. One component is piecewise smooth and models object shapes. Another component consists of the texture in the image, possibly including some noise. Image decomposition is useful for a host of image processing tasks, e.g. texture segmentation [1] and image inpainting [2]. Here, we consider ways of improving the quality of image decomposition using the basic variational approach of Meyer [3] by adding extra regularization terms, so that when decomposition is used for other image processing tasks, such as those noted above, better results can be obtained. A measure of quality of image decomposition found in the literature [4] is the absence of cartoon edges in the texture component of the decomposition. In this paper, a method called Improved Edge Segregation (IES) image decomposition is introduced, which ensures this quality measure is high. A very efficient parallel implementation of IES, suitable for multi-processor systems, is also described.