Distortion-constraint compression of three-dimensional CLSM images using image pyramid and vector quantization
The confocal microscopy imaging techniques, which allow optical sectioning, have been successfully exploited in biomedical studies. Biomedical scientists can benefit from more realistic visualization and much more accurate diagnosis by processing and analysing on a three-dimensional image data. The lack of efficient image compression standards makes such large volumetric image data slow to transfer over limited bandwidth networks. It also imposes large storage space requirements and high cost in archiving and maintenance. Conventional two-dimensional image coders do not take into account inter-frame correlations in three-dimensional image data. The standard multi-frame coders, like video coders, although they have good performance in capturing motion information, are not efficiently designed for coding multiple frames representing a stack of optical planes of a real object. Therefore a real three-dimensional image compression approach should be investigated. Moreover the reconstructed image quality is a very important concern in compressing medical images, because it could be directly related to the diagnosis accuracy. Most of the state-of-the-arts methods are based on transform coding, for instance JPEG is based on discrete-cosine-transform CDCT) and JPEG2000 is based on discrete- wavelet-transform (DWT). However in DCT and DWT methods, the control of the reconstructed image quality is inconvenient, involving considerable costs in computation, since they are fundamentally rate-parameterized methods rather than distortion-parameterized methods. Therefore it is very desirable to develop a transform-based distortion-parameterized compression method, which is expected to have high coding performance and also able to conveniently and accurately control the final distortion according to the user specified quality requirement. This thesis describes our work in developing a distortion-constraint three-dimensional image compression approach, using vector quantization techniques combined with image pyramid structures. We are expecting our method to have: 1. High coding performance in compressing three-dimensional microscopic image data, compared to the state-of-the-art three-dimensional image coders and other standardized two-dimensional image coders and video coders. 2. Distortion-control capability, which is a very desirable feature in medical 2. Distortion-control capability, which is a very desirable feature in medical image compression applications, is superior to the rate-parameterized methods in achieving a user specified quality requirement. The result is a three-dimensional image compression method, which has outstanding compression performance, measured objectively, for volumetric microscopic images. The distortion-constraint feature, by which users can expect to achieve a target image quality rather than the compressed file size, offers more flexible control of the reconstructed image quality than its rate-constraint counterparts in medical image applications. Additionally, it effectively reduces the artifacts presented in other approaches at low bit rates and also attenuates noise in the pre-compressed images. Furthermore, its advantages in progressive transmission and fast decoding make it suitable for bandwidth limited tele-communications and web-based image browsing applications.