Myagmarbayar Nergui et al.

Fig. 11.3. (a) An original retinal fundus image and (b) a watermarked/embedded retinal fundus image.

watermarked retinal fundus image of Fig. 11.3b is visible to the naked eye. Every pixel of the retinal fundus image comprising color data (red, green, blue) is represented by 24 bits.

11.3. Compression Technique

Compression techniques compress data so that it can travel on less bandwidth and can be stored using less memory. Compression schemes improve throughput over standard channels, and is commonly performed on digital medical images to reduce redundancy. Most compression algorithms use the repetition contained in data. For instance, a character set that includes letters, digits, and punctuation is normally composed of a seven-bit ASCII code, but a compression algorithm can use a three-bit code to represent the eight most common letters. The compression ratio is equal to the number of bits before compression over the number of bits after compression.

Two important compression concepts are lossless and lossy compression, as explained below.

In lossless compression, data is compressed without any loss of data, and all the data is available for use. Lossless compression is necessary for many medical images. Lossless compression is also known as entropy coding since it uses statistics/decomposition techniques to remove redundancy, and is applied to medical imaging.

The following techniques are included in lossless compression:

•Run length encoding,

•Huffman encoding,

326

Reliable Transmission of Retinal Fundus Images

•Lempel-Ziv-Welch (LZW) coding, and

•Arithmetic coding.

Lossy compression standards are JPEG (Joint Photographic Experts Group), MPEG (Motion Picture Experts Group), and MP3 (MPEG-1, Layer 3). Some loss of information is acceptable in the lossy compression technique. In these cases, it is acceptable to lose some data information to create smaller graphics files. The loss could be applied as a color resolution or as a graphic detail. For instance, high-resolution information points could be lost whenever a picture is displayed on a low-resolution instrument or equipment. The loss is also satisfactory in voice, audio, and video compression, depending on the desired quality. Lossy algorithms supply higher compression ratios than lossless algorithms do. Lossy algorithms are used for applications that can work well with lower quality reconstructed images. Using this scheme, the decompressed image is not identical to the original image, but is reasonably close to it. The loss of information in the image is unperceivable to the human eye.

Lossy compression can allow for compression ratios of between 100:1 and 200:1, relying upon the type of data being compressed. However, the lossless compression ratios commonly accomplish only a 2:1-compression ratio.

The following techniques are included in lossy compression:

•Transformation coding,

•Vector quantization,

•Fractal coding,

•Block truncation coding, and

•Subband coding.

Lossless compression standards are Gzip, Unix compress, zip, GIF, and Morse code.

The lossless image compression occurs in two stages: decorrelation and entropy coding. Decorrelation is a run-length coding technique, which removes spatial redundancy or inter-pixel redundancy. The second stage takes out coding redundancy. This stage includes Huffman coding, LZW, and arithmetic coding.

327