The use of difference color models with integer and half-integer coefficients for compression of images in modified jpeg graphic format

In the article a method and proper algorithms of reducing the size of compressed images and accelerating decoding in a modified JPEG format by using alternative difference color models with integer and half integer coefficients instead of the color model YCbCr are suggested. The use of the proposed difference color models reduces the size of each individual images due to intercomponent decorrelation and accelerates the decoding by using operations with integers and bitwise operations instead of operations with float numbers or scalable integers. Main findings of the study:
1. There is no universal color model that could allow to reach optimal decorrelation between components for all the types of images. Even for images of the same type, different color models may be optimal (in the sense of decorrelation).
2. In the graphic formats it is expedient to use discrete color models with integer or halfinteger coefficients for compression of individual images, instead of color models with valid coefficients, if they predict the decrease of compression coefficient. On average, such color models accelerate decoding by 3%.
3. For the image compression without losses, it is worthwhile to use difference color models with integer coefficients, and for the image compression with losses, it is better to use discrete color models with half-integer coefficients.
4. While compressing images with loses, color models with half-integer do not increase the range of possible values of individual components and do not significantly affect the image quality. If you want to minimize RMSE, then carrying component in a difference color model should be formed from the half sum of other components. If it is necessary to accelerate decoding as much as possible, then it is expedient to form the carrying component in a difference color model from one of the input components.