Myagmarbayar Nergui et al.

to assure data retains integrity, as it passes through imperfect transmission media. We have shown the outcomes found by applying turbo codes in this study in Fig. 11.10.

11.5. Results

After patient information is encrypted by using AES, it is then hidden in the retinal fundus image using digital watermarking; the image with hidden patient information is compressed using the lossless compression technique. In this study, Huffman coding technique is applied for lossless compression technique, which is given a 0.736-compression ratio. Then, the compressed watermarked image is changed over a bitstream and is encoded by employing suitable ECCs. In this chapter, we assume that the channel modulation is the binary phase shift keying (BPSK). Then modulated signals are passed through an assumed imperfect transmission media, which is corrupted by AWGN. The ultimate goal of our study is to quantify the character of the recovered text data and quality of rebuilt retinal image. The main quantity is that the BER is measured as a function of the SNR after decoding the obtained retinal image corrupted by the communication channel noise. Figures 11.5 (a–i) show us to visually estimate the difference between the images (also patient information) as retrieved over an AWGN channel (9- dB SNR) and the recovered image (also patient information) after being worked out by the (15, 11) Hamming code, (15, 11) BCH code, and (15, 11) RS code. From these figures, we can see that the quality of images and text data has improved significantly after applying ECCs. In this study, five types of ECC schemes are applied for the robustness and reliability of the transmission system. That is the Hamming code, the BCH code, the RS code, the convolutional code, and turbo codes. Then, we have compared the performance of Hamming code (15, 11), the BCH code (15, 11) and the RS (15, 11) code.

We show the original retinal fundus image in Fig. 11.5a, the original patient text information that has to be hidden into retinal image in Fig. 11.5b, the image after hiding text data in Fig. 11.5c, respectively. The watermarked image (image plus text) corrupted by noise with 9-dB SNR as received at the receiver without ECC is shown in Fig. 11.5d. The reconstructed image

336

Reliable Transmission of Retinal Fundus Images

Fig. 11.5. (a) An original retinal fundus image, (b) original patient text information, (c) the retinal fundus image after hidden text data, (d) the watermarked retinal fundus image (image + text) corrupted by noise with 9-dB SNR without using ECC, (e) the reconstructed retinal fundus image using the (15, 11) Hamming code for 9-dB SNR, (f) the reconstructed retinal fundus image using the (15, 11) BCH code for 9-dB SNR, (g) the reconstructed retinal fundus image using the (15, 11) RS code for 9-dB SNR, (h) the recovered patient information corrupted by noise with 9-dB SNR without using ECC, and

(i) the recovered patient information using RS code for 9-dB SNR.

using the (15, 11) Hamming code (9-dB SNR) is shown in Fig. 11.5e. The reconstructed image using the (15, 11) BCH code (9-dB SNR) is shown in Fig. 11.5f. The reconstructed image using the (15, 11) RS code (9-dB SNR) is shown in Fig. 11.5g. The recovered patient information corrupted by noise with 9-dB SNR as received at the receiver without ECC is shown in Fig. 11.5h. The recovered patient information using RS code (9-dB SNR) is shown in Fig. 11.5i. In this chapter, we used the (15, 11) Hamming code

337

Myagmarbayar Nergui et al.

the (15, 11) BCH code, and the (15, 11) RS code over GF(24). These three codes have same code rate, which is 11/15 = 0.7333. Nevertheless, their error-correcting capabilities are different. In this study, we wanted to compare with the results of these codes. The (15, 11) Hamming code and (15, 11) BCH code can correct up to one bit in error within a 15-bit length codeword.

However, the (15, 11) RS code can correct up to two symbols in a 15-symbol codeword. In a fading channel, errors tend to occur in bursts rather than in random patterns. Interleaving, the rearranging of the symbol sequence, increases the burst error-correcting capability of an ECC. A deinterleaver is the inverse process of interleaver. A block interleaver is very simple and widely applied in the communication systems. In this chapter, in order to improve the capability of error correction, a block interleaver is applied along with the (15, 11) Hamming, the (15, 11) BCH, and the (15, 11) RS codes.

The plots in Fig. 11.6 show the compared result of performance of the (15, 11) Hamming code, the (15, 11) BCH code, and the (15, 11) RS code on transmission of retinal fundus image. From this result, we can observe that this RS code performs better than the Hamming and BCH codes, and has a good CG with 4 dB at a BER of 5 × 10−6. From this plot, the BER of uncoded case will reach to roughly 10−3 at SNR of 9 dB. This raised BER indicates almost one bit in error for every 103 bits passed through. Hence, using the RS code leads to increased information integrity. The RS code results are similar to those found in literature. Wicker has given a plot of the probability of the decoder error (essentially the SER) as a function of SNR for a (31, 27) two error-correcting ECC.23 It is observed that the SER is approximately 10−5 at an SNR of 8 dB.

We use a Huffman coding compression technique after the watermarking technique. Huffman coding is given a 0.736-compression ratio. We use Hamming, BCH, and RS codes with a code rate of 0.733 as error-correcting codes. This compression ratio is similar to the code rate used in ECC, meaning that data size after using ECCs is equal to original image size. We save the memory space of storage media and transmission in channel.

We check the performance of these codes over time variant channels. We assume that this channel can change suddenly from a good state (channel introduces few errors) to a bad state (channel introduces many errors). The

338

Reliable Transmission of Retinal Fundus Images

Fig. 11.6. Compared result of (15, 11) Hamming code, (15, 11) BCH code, and (15, 11) RS code performances with block interleaver (The retinal fundus image passed through noisy channel which is corrupted by AWGN channel.).

bad state errors are related to burst errors in fading channels. For insuring the performances of Hamming, BCH, and RS codes over such channels, we randomly alter the SNR on the communication channel. Such channel SNR values are altered over 10 values roaming amongst 10 dB (a large amount of SNR) and 0 dB (a small amount of SNR). The performances of the Hamming, BCH, and RS codes are recorded in detail of the BER plot in Fig. 11.7. We find an improvement of about 5 dB at a BER of 10−5 for RS code.

We can see that (15, 11) RS code is more beneficial fitted for burst errors correction than (15, 11) Hamming code and (15, 11) BCH code.

Figure 11.8 shows the performance of convolutional code with a code rate of 0.5. This convolutional code performance is better than the Hamming and BCH code performances but less than the RS code. From these results, we can conclude that the RS code is a better error-correcting code capability; error occurs in random and burst.

339