A perceptually perfect image resizing scheme

In a numerous image applications, image resizing is becoming indispensable as it impacts the bandwidth and storage requirements tremendously. The image resizing process introduces (or eliminates) many pixels to (or from) the original image. The values of such pixels and their positions must be carefully determined otherwise visible distortion and significant degradation in the quality of the image may occur. Several methods have been employed to resize images including pixel replication; linear interpolation; higher order interpolations, Bezier methods, DCT-based, wavelets and others. We introduce a new perceptually perfect image-resizing scheme that near optimally preserves edges and highly maintains the quality of homogenous regions. In this technique, the image is segmented via an efficient edge detector to produce an edge image and independent homogenous regions. The edge image is resized separately from the homogenous regions via chain coding and elaborate look-ahead-and-back tables technique. Homogenous regions are resized using a merciful adaptive region-based interpolation that exploits the characteristics of each region. At the end, the two parts are summed up to produce the desired resized image. Simulation results of numerous test images show that the proposed technique is subjectively and objectively far better than published results.