Page 137 - 2023-Vol19-Issue2

133 | Al-Furaiji, Tsviatkou & Sadiq

Fig. 4. Dependences of the mean error on the spectral channel number: (a) for pixel inter-channel correlation;
(b) for bitwise inter-channel correlation (12th bit plane); (c) for bitwise inter-channel correlation (5th bit plane);

(d) for bitwise inter-channel correlation (0th bit plane)

number of spectral channels; CAC > 1 – is a coefficient taking IV. CONCLUSION
into account the computational (time) complexity of the arith-
metic coder compared to RLE (operate at frequency Y X f ); f In this paper, a combined codec structure for compressing
– is clock frequency f = N fI; fI – is HSI formation frequency. bit planes of images without loss in the spatial domain based
It is found, experimentally, that CAC ˜ 11. on arithmetic and RLE coders is proposed. The efficiency
of combined coding for compressing channel differences of
The structure of HSI coder also includes spectral channel hyperspectral images is established. The choice of optimal
multiplexers, operating at a frequency f at the control inputs combinations of bit planes for arithmetic and RLE coding
and at a frequency 16Y X f at the data inputs; a multiplexer of allows to increase the compression ratio of the differences
codes of bit planes of differences of spectral channels, operat- of neighboring HSI channels by an average of 1.4 and 3.1
ing at a frequency 10 f ; a multiplexer of codes of bit planes times compared to arithmetic and RLE coding, respectively.
of reference channel, operating at a frequency 5 f / (N - 1); a A rule has been developed for combined coding of differ-
multiplexer of reference codes and differences of HSI chan- ences of adjacent channels of hyperspectral images, using
nels, operating at a frequency f / (N - 1). In a sequential arithmetic coding for the upper sign plane and bit planes 8–2,
implementation, the computational complexity of the com- RLE coding for bit planes 14–9, and transfer without coding
bined coder is about 2Y XCAC f + 6Y X f at a large number of for bit planes 1–0, which made it possible to increase the
spectral channels, which is about 2.5 and 28 times more than compression ratio of the differences of neighboring channels
arithmetic and RLE-coding, respectively. of hyperspectral images on average 1.2 and 2.6 times com-
pared to arithmetic and RLE coding, respectively, with an
Figure 7 shows the structure of combined HSI decoder. increase in computational complexity of 2.5 and 28 times.
Based on this rule, the coder structure of hyperspectral images