Reducing Boundary Errors in Wavelet Image Compression
Dana Černá, Václav Finěk, Theodore E. Simos, George Psihoyios, Ch. Tsitouras · AIP conference proceedings · 2010
Originally, the discrete wavelet transform was designed for an infinite input data. Applying discrete wavelet transform directly to images in lossy wavelet image compression leads to large errors and artifacts at the tile and the image boundaries of the reconstructed image. To reduce boundary errors and artifacts the image is usually first extended and the discrete wavelet transform is applied to the extended image. An alternative approach consists of using directly the discrete wavelet transform with special filters near the boundary. In our contribution, we show that the tiling artifacts are significantly reduced and that the error of the compression is smaller with the discrete wavelet transform with recently derived boundary filters than with other methods using an extension of an image.