An Efficient Memory Model for Implementing Image Resizing Algorithms in a Distributed Environment

Image resizing algorithms are a classic case of algorithms involving local operations over a region of pixels in an image. The objective is to produce a reduced or enlarged image while maintaining original information content or minimizing the mean square error between corresponding pixels of original and resized images. Most resizing algorithms rely on pixel values within a pre-defined neighborhood of a pixel in the original image to compute pixel values in target images. High frequency or high energy pixel regions in an image are more prone to distortions/errors in the resized image. Content-aware algorithms minimize this impact at the cost of more computational complexity and cost. Parallel/distributed implementations of such algorithms require an efficient methodology of image data partitioning to minimize interdependency of the processing units and/or memory storage to avoid shared memory access bottleneck. A restricted shared memory model is described herein that is well-tailored for most of the computational techniques used in image resizing algorithms. Implementation results are described for some well-known algorithms that demonstrate the suitability of the model and its scalability to cater for large image sizes.