Wavelet compression of images.

Project Details

Description

The technological improvements in digital electronics and the significant cost reductions in digital imaging systems have triggered an increasing demand for handling images in digital form. Inevitably, digitised picture information exchange with increased bandwidth requirements has emerged: videophony, teleconferencing, tele-shopping, video-on-demand, tele-surveillance, tele-control, digital TV, medical data bases. Video communication has become an integrated part of computer systems having multi-media capabilities.All these applications require a source coding scheme, achieving both a very high compression ratio (CR) for communication channels with low bandwidth (PTSN, mobile telephony and basic rate ISDN) and a controllable CR for the adaptation to shared channels with a variable bandwidth (e.g. ATM, LAN). It is well known that besides Region Oriented Transform Coding (ROTC), wavelet based subband coding is a good candidate to implement these requirements. The well-known standard compression methods, such as JPEG or MPEG, generate unacceptable block artefacts at high compression ratios. Image coding through wavelet transforms fundamentally solves this problem. Moreover the wavelet transform is well localised in both space and frequency domains and by its orthogonality eliminates the space correlation. The application of this approach is additionally motivated by its similarity to the processes in the human visual system, therefore a good visual quality may be preserved at high CR. Wavelet based compression allows an easy performing of progressive transmission and it is expected that efficient implementations of real time generic wavelet based image and video codecs will appear sooner than those for ROTC. Starting from the theoretical aspects of wavelet based coding, several fundamental problems have to be solved to bridge the gap between theory and application. The development of a lossless wavelet based compression scheme suitable for progressive transmission of images is a prerequisite for medical applications like tele-diagnosis and fast browsing through image databases. Moreover, evaluation of the strategies for bit rate control, new approaches for the quantization of the wavelet coefficients like the selective quantization of the subbands, and filters design are but few concepts to be studied. An image coding technique that exploits the similarities among blocks in the different subbands of a wavelet representation of the image seems promising, reducing further the quantization cost.For the compression of time sequences of images, the main temporal redundancy to be exploited to get good compression results is the one due to the motion of objects in the images. Going from full-search, over logarithmic search towards hierarchical motion estimation, calculations indicate a significantly decreasing processing load. Therefore appropriate algorithms for hardware implementation are obtained treating the motion vectors on a multiresolutional basis using a wavelet pyramid. Furthermore, an interesting path to look at is the possibility of coding the motion vectors guided by a contour tracking. Exploiting the facts that the redundancy of neighbouring motion vectors, at the lowest resolution level, is high at the contours, and that the higher frequency components are rather small within the image segments, a coding gain should be feasible.Zerotree coding of grey value images is a very efficient method based on the separate coding of non-significant and of significant bits of the wavelet coefficients of the image. The extension towards colour images seems promising and high compression ratios preserving the visual quality of the reconstructed image can be achieved. The initial idea that will be worked out in the project is briefly described hereafter. The wavelet transforms for the red, green and blue components of the colour image are computed separately. The bits of the obtained wavelet coefficients are separated in common non-significant bits (i.e. bits that are insignificant for the red coefficient, for the blue coefficient and for the green coefficient) and into separate colour significant bits. The common non-significant bits are coded in a dominant pass exactly as for the gray value case. The other bits are coded in three separate subordinate passes, i.e. a red subordinate pass, a green subordinate pass and a blue subordinate pass.The algorithmic development should consider the particular requirements imposed by efficient implementation. IMEC-VUB have started developing design guidelines for wavelet based subband codecs and the complementary algorithmic ideas of UPB-VUB will be used to improve both the design tools and the codec systems.
AcronymVLW11
StatusFinished
Effective start/end date1/10/9719/06/00

Flemish discipline codes

  • Electrical and electronic engineering

Keywords

  • Region Oriented Transform Coding - ROTC
  • Wavelet
  • Compression methods