Data Dependent Compression Based on Reconfigurable DCT Using JND Algorithm

This paper presents a data dependent compression algorithm for applications using discrete cosine transform (DCT). The algorithm can efficiently trade off image quality and computation complexity. Based on the just noticeable distortion (JND) of four reference positions, two work modes can be switched for different frequency components to reduce computation complexity. To support this algorithm, we also propose a reconfigurable DCT architecture based on the CORDIC algorithm. The proposed architecture can evaluate N point DCT as well as N/2 point DCT without change any architecture. Experimental results show that the proper threshold can be determined by trading off the computation complexity and image quality, which will save about 40% computation compared to conventional operation at the expense of minor image quality degradation.