Image/video compression based on bracket classified coding and optimal joint coordinate motion estimation

Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


D. J. Pease


image compression, algorithm, Electrical engineering, Computer science

Subject Categories

Computer Engineering


The need for an efficient image coding scheme continues despite very significant advances in the area of image processing. Most image coding schemes are either adaptations of data compression functions, such as discrete cosine transform (DCT) or vector quantization, or utilizations of statistical properties of images. This research describes a new image coding scheme to be used in transform coding. This scheme adds a computationally simple image classification pre-processor to convert images to new and useful forms that are easily compressed.

Based on spatial image features, a proposed classifier divides non-overlapping pixel groups called brackets into classes, each of which is identified by a label. Only a subset of computed pixel values are sent for DPCM/DCT encoding. Exploiting close correlation of class labels, a novel scanning method to generate drives is developed. A drive represents a run of labels with the same class. It is the drive, rather than individual labels, that is coded using Huffman algorithm. The proposed scheme using a 2 x 2 bracket segmentation achieves excellent visual quality, bit rate, and processing speed.

Block matching motion estimation/compensation has emerged as a very efficient technique to remove temporal redundancies in video signals. Based on a distortion function, it searches for the best match between a block of pixels in the current picture frame and a number of blocks in close proximity in the previous frame. Most of the published algorithms sequentially reduce the search area surrounding the optimum match at each search step. This research describes a novel methodology to advance the search towards the joint coordinate of the two optimum matches in each step. A new algorithm is built on this methodology to avoid redundant searches.

Our algorithm relies on a statistically valid assumption of the distortion function as a convex function with elliptical contours. The algorithm achieves quick convergence and precise estimation of motion estimation. In addition, two adaptive approaches are proposed to improve the efficiency of motion estimation. They adapt to variations of block displacements by varying the size of the initial search region. The close correlation between block displacements is exploited in achieving this object.


Surface provides description only. Full text is available to ProQuest subscribers. Ask your Librarian for assistance.