This paper introduces a new view onto the problems of astronomical image data compression. The idea is to find precise models of real objects from image and describe the image with the help ofthem. This approach should lead to extreme compression ratios but maintain accurate astronomical information and in most cases also visually lossless image quality. Lossless compression is possible at the expense of low compression ratios but these ratios should be still able to compete to recent compression algorithms used in astronomy (HCompress, PMT or FITPRESS). The core ofthis method is very precise astrometry and photometry measurement, the rest of the image, which will not fulfill the expectation of our model, can be coded using wavelet transform with optimized mother function.