Integrated Circuits and Systems group, IIT Madras

Design and Implementation of an Efficient Progressive IMage transmission System using Pruning Algorithms and a parallel Architecture.

By S.Venkatesh

Abstract

In applications involving image communications, resolution if images and channel speeds restrict fast transfer of images over networks. Progressive Image transmission(PIT) has been found to be an ideal candidate for such applications involving image communications. In this work, an implementation of a PIT system using a combination of pruning algorithms and parallel architecture using a DSP and a host processor in PC has been designed and implemented.

Pruned FCT algorithms have been used to enhance the speed of the system further. In order to implement the pruned FCT algorithms on a DSP, a modified FCT butterfly structure has been proposed in this work along with an efficient method of implementation of the pruning algorithm on DSP, thereby eliminating the overheads present in the pruned FCT algorithm.

In the PIT system designed in this work, a guaranteed minimum quality is being ensured in the first progression of any image using adaptive pruning algorithm. The pruning level prediction method used in the adaptive pruning algorithm has been simplified further to enhance the speed of the system. The savings in operation obtained using pruning algorithms have been extended to the quantisation and the Huffman coding steps following the DCT step in JPEG.

A data dependent inverse FCT has also been proposed in this work to extend the advantages obtained using pruning algorithm at the encoder to the decoder. A differential pruning level method has been proposed to impart the knowledge of pruning level employed at the encoder to decoder.

With efficient distribution of the execution of the different steps in JPEG between a DSP and a PC and with the use of pruned FCT algorithm, the entire PIT system has been found to achieve an increase in speed by a factor of 2.8 compared to the normal implementation of the progressive mode of JPEG on a general purpose processor.