Eye-movement controlled image coding

Sammanfattning: This thesis proposes eye-movement controlled image coding schemes. An application is found in interactive systems where a human operator controls a vehicle at a distance. Such a system would include an eye-tracker to pick up the point of regard of the viewer. The coding schemes can then utilize the fact that the acuity of the human visual system is lower in the peripheral vision than in the central part of the visual field. This offers great potential for data compression.The thesis consists of four parts. First a visual acuity based model is proposed. It translates a minimum angle of resolution function describing the visual acuity to a corresponding limiting size or frequency in a displayed image on a flat screen. Secondly  a polar down-sampling based coding scheme is proposed. This method provides large data compression. However, it can result in visual artifacts as a result of the down-sampling and reconstruction. To avoid these artifacts a coding scheme based on pre-filtering is proposed in the third part. This method preserves the original sampling grid. The result is better visual quality than for the previous scheme.In the fourth part of the thesis a method to incorporate the visual acuity model into transform coders is suggested. Transform components related to details in the image which cannot be perceived by the observer are set to zero. The result is a scheme with relatively low complexity which provides high data compression while the visual degradation is negligible. This method is applied to different kinds of transform coders and their results are discussed.Visual experiments have been performed to evaluate the results of the proposed schemes. Experience from these experiments are presented and discussed.