The success of image processing (IP) applications implemented on Instruction Level Parallel (ILP) RISC systems relies on the efficient utilization of the CPU resources. While in general it is the compiler that organizes the machine code in such a way to maximize the hardware resource utilization it was recently shown that in the case of IP programs compilers do not actually produce efficient machine code. Therefore special coding schemes (based for example on loop unrolling) must be adopted to help compilers to produce efficient code. This paper investigates some of these special coding schemes. More specifically it describes the effects of different loop unrolling schemes and introduces a new source program transformation based on the dynamic scheduling of the program iterations which, under some hypotheses, is shown to be more efficient than unrolling. A case study on the implementation of the Hough Transform is presented to validate the results of this paper.