An Algorithm for Grayscale Images Compression based on the forward Adaptive Quantizer Designed for Signals with Discrete Amplitudes

Abstract

In this paper an algorithm for grayscale image compression is presented, based on implementation of forward adaptive quantizers designed for signals with discrete amplitudes. Experiments are done, applying this algorithm on standard grayscale images and obtained results show that significant reduction of the bit-rate can be achieved (about 40%), maintaining very high quality of the reconstructed image, i.e. near lossless compression is performed. Ill. 2, bibl. 6, tabl. 2 (in English; abstracts in English and Lithuanian).

Coding algorithm for grayscale images based on Linear Prediction and dual mode quantization

“…As a measure of quality of the compressed image is commonly used PSQNR (peak signal-to-quantization-noise ratio), while the total average bit-rate is used as a measure of compression quality. Obtained results show that by using the proposed model for grayscale image compression, compared to the BTC model published in Savic et al (2012aSavic et al ( , 2012b, achieved gain is equal to 2.845 [dB] for N = 32. If N = 16, achieved gain is 6.140 [dB].…”

“…From the above described algorithm it is known that T h2 is a decision making threshold that defines the range of usage of offered quantizers. Related to that, we will obtain representational variances for ranges [0, T h2 ] and [T h2 , 255] by using following equations, respectively: The piecewise uniform quantizer Q 1 has the maximal amplitude x max1 ¼ k 1r1 and it will be realized with N levels grouped in L regions (Savic et al 2012a(Savic et al , 2012b. Each region has M = N/L uniform output levels.…”

“…Next, obtained numerical results of applying the proposed model are analyzed in Section 2.1. Afterwards, in Section 2.2 are shown numerical results for classic BTC model and the comparison of the proposed compression algorithm to BTC model presented in Savic et al (2012aSavic et al ( , 2012b. Furthermore, appropriate comparison to DP-DPCM model presented in Wu (2014) is provided.…”

“…Piecewise uniform quantizer incorporates good characteristics of the both uniform and nonuniform quantization (Savic et al, 2012a(Savic et al, , 2012b. Quantizer range is divided into k segments (usually k = 2 n , n 2 N) and uniform quantization is applied in each one.…”

“…In this paper we propose an algorithm that includes the both uniform and piecewise uniform quantization and uses a pixel value prediction as an improved method of grayscale image compression (Jayant & Noll, 1984;Peric, Petkovic, & Dincic, 2009;Savic, Peric, & Panic, 2011;Savic et al, 2012aSavic et al, , 2012b. Dynamic range of the source output is considerably larger than the dynamic range of the differences between samples.…”