In advance of the image capturing technology, large amounts of similar images are created. Instead of compressing each similar image individually, compressing the image sets by removing the inter-image redundancy would reduce the storage and the transmission time. The video encoding technology assumes a temporal correlated image sequence while the image set compression need to accept the less restricted images in the set. Random access in decoding a specific image is another criterion to make the image set compression practical. However, only a few methods were proposed to deal with the problem.

In this thesis, a new lossless compression method was derived from a theoretical model by extracting the low frequency in an image set. In the observation, two blurred similar images look like...[ Read more ]

In advance of the image capturing technology, large amounts of similar images are created. Instead of compressing each similar image individually, compressing the image sets by removing the inter-image redundancy would reduce the storage and the transmission time. The video encoding technology assumes a temporal correlated image sequence while the image set compression need to accept the less restricted images in the set. Random access in decoding a specific image is another criterion to make the image set compression practical. However, only a few methods were proposed to deal with the problem.

In this thesis, a new lossless compression method was derived from a theoretical model by extracting the low frequency in an image set. In the observation, two blurred similar images look like each other. The blurred images could be regarded as the low frequency components of the images. In our model, a low frequency template is created and used as a prediction for each image to compute the differential image. And intra-prediction has been performed on each differential image. In the last step, the residue has been compressed by the standard lossless compression methods. Enhancements have been proposed to compress the low frequency template. In comparing with other set redundant methods, this model proves the reduction in the entropy and hence the bit rates. Experiments were conducted on magnetic resonance brain image sets and computed tomography brain image sets from different patients. Results shows there were up to 30% gain over the existing methods and the proposed method performed better in smaller image sets. In the future, the proposed method could be extended to colour image sets and lossy compression.

Index Terms— Image compression, set redundancy compression