First Absolute Moment and Variance Estimators Used in JPEG Reconstruction
Robert Krupiński, Jan Purczyński · IEEE Signal Processing Letters · 2004
Rice and Rabbani modeled the distribution density of discrete cosine transform (DCT) coefficients by means of the Laplacian distribution. Parameters of the model were estimated on the basis of quantized values of DCT coefficients available to the decoder, employing the maximum likelihood method (MLM). In this letters, two estimators (alternatives to the maximum likelihood one) of the discrete Laplacian distribution modeling DCT are presented. Reconstructed values of images are adjusted based on a distribution with estimated parameters. The reconstruction of DCT coefficients, similarly as in [J. R. Price, M. Rabbani (1999)] and [R. L. de Queiroz (1998)], is carried out by application of centroids. For selected images, the values of PSNR coefficients and values obtained through the MLM are compared.