20-06-2011, 03:41 PM
[attachment=14170]
A Pseudo Lossless Image Compression Method
I. INTRODUCTION
Data compression methods can be reversible (error-free) or irreversible (lossy).
A reversible scheme can only achieve a aximum compression ratio of about 2.5, but will allow exact recovery of the original image from the compressed version.
This compression ratio is limited by the noise in the image which degrades the correlation between pixels.
I. INTRODUCTION(cont.)
Today, lossy compression methods are not being used by radiologists in primary diagnoses because radiologists are concerned with the legal consequences of incorrect diagnosis based on a lossy compressed image.
In this study, it is proposed a new image compression technique called “pseudo” error-free compression.
I. INTRODUCTION(cont.)
The technique comprises three steps:
(1) estimating the noise level of each pixel,
(2) identifying the noise bits and modify them,
(3) compressing the image reversibly.
The idea behind this method is that we can modify the noise bits in the image without affecting the quality and yield better compression ratio.
I. INTRODUCTION(cont.)
The advantage of this new technique is that the compression ratio is higher than reversible compression method and the image quality is not affected.
II. MATERIALS AND ETHODS
When an image is contaminated by noise, the correlation between pixels is degraded. As a result, the efficiency of image compression is decreased.
The signal bits contain image structural information. The noise bits are contaminated by noise.
II. MATERIALS AND ETHODS
A.Estimation of noise level
The signal component image is obtained by median filtering the original image with a 3×3 window.
Define a residual image as the subtraction between the original and its signal component images. The residual image contains structured edges and noisy background information.
Standard deviation can be used to estimate the noise for the smoothed parts of the image.
B.Noise bits and zeroed image
The number of noise bits can be estimated by simply taking the logarithm of the noise level.
Since the noise bits do not contain any structural information, they can be modified without deteriorating image quality. In this study, all the noise bits of an image are set to zero.
Performing a reversible compression on these images will yield better compression ratio than on the original images.
C.Reversible image compression
A common characteristic of radiological images is that neighboring pixels have a high degree of correlation.
In this study, decorrelation is performed by the subtraction between adjacent pixels. The subtraction is a one dimensional version of the differential pulse code modulation (DPCM).
C.Reversible image compression
Arithmetic coding assigns a code word to each symbol an interval of real numbers between 0 and 1. It exploits the distribution of the image histogram, by assigning short intervals to the most frequently occurring amplitudes and longer intervals to the others.
III. RESULTS
III. RESULTS(cont.)
III. RESULTS(cont.)
III. RESULTS(cont.)
IV. DISCUSSION
Our method modifies the noise bits of the data. The loss in the data is purely noise and will not cause any loss in visual and information.
For signal estimation, window size needs to be small in order to save detailed information. For standard deviation calculation, the window size should be large enough for statistically meaningful.
IV. DISCUSSION(cont.)
The residual image, whose signal part can be effectively removed, is the right choice for correctly measuring the noise bits.
The mean filter can smooth the images better than median.
The qualities of processed images are the same as compared with those images by loosy JPEG2000 image compression at compression ratio around10.