Adapting transform coding to texturally segmented images

Nonstationarity in image textures requires implementation of variable-size partitioning into stationary regions for coding purposes. The popular methods in regular decomposition fail in accurately following the various region boundaries. In the present work, the boundaries of different textures in the image are detected using a one-dimensional model of each row of the image. The autoregressive algorithm operates in 8 directions to enable an efficient detection of the boundaries. The edge map is then achieved by taking the result of edge detection operation in all directions into account. Transform coding and run-length coding are applied to encode the textures and the boundaries, respectively