In this case, the JPEG-LS GR code consists of a 47-bit prefix, a 1-bit separator, and a 16-bit suffix; therefore, the replaceable JPEG-LS GR prefix is 48 bits for the 16-bit-depth images.
Either GR or JPEG-LS GR code is used for each codeword when the regular mode is utilized, and this regular mode is the sole focus of this paper.
First, the replacement of the replaceable JPEG-LS GR prefix with location information can reduce the codeword size.
This paper identifies the possibility that the JPEG-LS GR codeword can still be shortened using a simple method; that is, the location information is used instead of the replaceable JPEG-LS GR prefix.
The location map consists of the simple (row, col) coordinates of such JPEG-LS GR code in this paper.
At a position (row, col) = (339, 211) of the 512x512 image in Table 1, for example, the JPEG-LS GR code is [00000000000000000000000|1|01101110.
When the replaceable prefix is encountered for a pixel that is quantized by eight bits, the manifestation of the leading 23 zeros means that this sample is encoded by the JPEG-LS GR method, whereby the value of the quotient q [greater than or equal to] 23 and the suffix value is [N.
The performance of the proposed method was compared with that of the JPEG-LS algorithm (see Tables 4 to 8).
The exact output bitstream size of JPEG-LS depends on two kinds of zero padding.
For the lena image (256x256), the original GR and JPEG-LS GR output stream size without any padding is 299,901 bits, and 123 bits are added through the padding process to make 300,024 bits; notably, since 300,024 is a multiple of eight, the first type of zero padding is not necessary.
2)]") are added to the pure JPEG-LS output of 299,901 bits; alternatively, 160 bits are reduced by the occurrence gain due to the 20 replaceable prefixes.
As a result, the output size of the JPEG-LS and the proposed method are 282,488 bits and 281,576 bits, respectively.