The current research in many areas of image processing involves operations performed in the compressed domain [Smith and Rowe 1993, Chang 93]. Compressed domain may be DCT, Wavelet, JPEG or MPEG etc. The algorithms for .these transformations may be used for information filtering such as feature extraction and edge detection. The image transformations such as compositing, occluding, and scaling may also be performed in the compressed domain. It is desirable to perform these operations on compressed data directly because the smaller size of data involves less computational complexity. The computation in the compressed domain eliminates the overhead of decoding the existing compressed information and then recoding it. Many times it is desirable to combine adjacent blocks in the compressed domain to perform feature analysis. A technique for DCT of a block composed of two halves of adjacent blocks was published in [Kuo & Fjallbrant 1991] and it was limited to signal processing in one dimension. This paper extends these ideas to image processing. The purpose of this paper is to present a new mathematical technique where a one dimensional vector block is not constrained to two halves of adjacent vectors and a two dimensional image block is not constrained to quarters of the four adjacent image blocks. This technique is elegant and adaptable to subblocks of any size. This technique is applicable all unitary transforms including DCT and Wavelet etc. The derivation of results and computational complexity of new algorithm are presented. © 1997 ACM.


Computer Science

Keywords and Phrases

Block processing; DCT; Frequency domain; Motion vector; Wavelet

International Standard Book Number (ISBN)


Document Type

Article - Conference proceedings

Document Version


File Type





© 2024 Association for Computing Machinery, All rights reserved.

Publication Date

01 Jan 1997