Design Of A Fast JND Algorithm Based On Discrete Wavelet Transform And Its Implementation

The improvement of digital watermarking technology makes it moreconvenient to protect the original multimedia information. The watermark basedon human vision system is widely used since it can both guarantee a robustwatermark and keeps a good quality of the watermarked images.Just Noticeable Distortion (JND) is a conventional algorithm to calculateHVS based watermark strength. But since its calculation is seriously complex,this technology is mainly used on computers but not embedded systems. In thispaper, conventional spatial domain JND algorithm is introduced by the analysisof simulation on Matlab. A novel fast JND algorithm is also proposed based onthe characters of discrete wavelet transformation. The novel algorithm can beeasily implemented by hardware description language to achieve a faster speed.Using JND as the quantization errors, the watermark is added to original bitimage by quantity index modulation to obtain the encrypted image. The algorithmproposed in this paper is then optimized to make the error of PSNR keeps below4%. To improve the speed of computation, a five-state pipeline architecture isproposed. The critical path of the circuit is reduced from Tsqrt+2Tm+7Ta toTsqrt/3.To verify the functional and timing of the proposed algorithm, it is modeledby Verilog and is embedded, and an SOPC system on FPGA is built to assist withthe core. An Altera EP2C35FPGA is used for the experiments. It costs1460logiccombinations and330registers to calculate a8-bit gray-level image. The systemruns well on120MHz clock frequency. The improved algorithm saved about75%computation time compared with the conventional JND method.