A Study on the Differential Compression Method for High-speed Transfer

Nowadays, most of people are using image for communication. And also, SNS (Social Network Service) which is used by young man have to be loaded images. So, we propose a new differential compression method for high-speed transfer. In this proposed method, the feature which is the encoding process according to subjectively image quality can be demonstrated. Human studies have shown that most of people have the interest about special area. Using the individuality, the relatively high quality can be obtained at the special area and the relatively low quality can be obtained at the other area in encoding process. In order to use these characteristics, the image is separated simple and complex block using 8x8 block standard deviation. And also, we have applied low compression rate and high compression rate to each region. The experimental results are demonstrated that proposed differential scheme can achieve better image quality at the high compression ratio. The proposed method is possible to maintain a high quality by differential compression of images. And then, we can apply to MPEC and H.26X.


Introduction
Image processing system is very important that can be encoding and decoding the original data using less then.Generally, digital image and video compression is mainly used JPEG based on DCT (Discrete Cosine Transform) technique 1 .JPEG is a common way on the whole area of image using the same compression ratio, but differential compressing method which use interest and non-interest region has been research.However, this differential compressing method which use pattern images and form information has the disadvantage computational complexity 2,3 .Thus, we propose a new compression method using Weber's law.Weber's law states that the justnoticeable difference between two stimuli is proportional to the magnitude of the stimuli on human perception.Additionally, it is possible to maintain a high quality by differential compression of images.

JPEG Compression
Whole image is divided into 8X8 blocks for JPEG compression.And then proceed with the DCT process for each block.DCT process is a process which the two-dimensional spatial data convert to spatial frequency information.
In other words, the pixel of the two-dimensional matrix devided into low-frequency and high-frequency components 4 .The reason for separation the image is to use the characteristic that human eye is more sensitive the low frequency components than high frequency components.On the other hand, the amount of data to be stored is reduced through the quantization process.However, the loss of data will be occurred.In addition, dividing the DCT coefficient to a large value, the amount of data increases and the loss of Geun-Ho, Lee, Sun-Woo, Ko and Soo-Tae, Kwon data becomes large.When we run DCT process to the 8×8 block, we can know that low-frequency area has a high sensitivity to recognize a difference.Therefore, a small value can be applied to the low-frequency area, and the highfrequency area is to use a large value for split.Currently, JPEG standard quantization coefficient has used the specifying table which determined experimentally in the ISO. Figure 1 is a standard quantization coefficient table.
Finally, the Huffman encoding is performed for data transmission 5 .The two-dimensional array of the 8X8 block is changed to a one-dimensional array through zigzag scanning.And then, data are encoded by using a table based on the number of zeros and non-zero data.However, JPEG compression method applies the same quantization table for every area.So, there is a problem that the region of non-interest takes a same data size with the region of interest area.

ROI (Rigion of Interest) Compression Method
In ROI compression method, the image is separated by two areas which is interest area and non-interest area.We can take a high quality for the region of interest by using a different compression ratio for each area.The RIO compression method has three different techniques.One is an edge-based technique, another way is a threshold-based technique and the other is an area-based compression technique.

Edge-Based Technique
This is a method to find the edge which means the boundary between the background and the object in the image.And then this area is set the ROI.Sobel, Robert, Prewitt technique are method to detect the contour using a gradient operator 6 .Canny mask is an application method using another mask operator when detects a contour.After removing the noise with Gaussian mask before the edge operating, performs other edge detection mask such as Sobel mask.

Threshold-Based Technique
This is a method to find the edge which means the boundary between the background and the object in the image.And then this area is set the ROI.Sobel, Robert, Prewitt technique are method to detect the contour using a gradient operator 7 .These methods can be identified by the method which divides the area by setting a single and multiple threshold value.

Area-Based Technique
This is the method which uses characteristics of adjacent pixels such as brightness, color and texture and so on.We can group contiguous area that has the same characteristics.There are area growth and clustering methods.Both the two methods is the way for clustering the similar pixels based on the reference point.So, this performance depends on the reference point which initially set.
As mentioned, conventional ROI method had an interest in only the object and background separation.However, human show an interest in not only the object but also color and pattern change area.So, the problem is that color and pattern change area is considered as a region of non-interest.

Proposed Compression Method
Figure 2 is a flow chart of the proposed compression method.After the DCT process, the ROI is setting by weber's law.And then, the ROI is compressed with differential quantization table

Weber's Law
Human vision is a non-linear reaction with respect to the intensity of the light.It is more sensitive in a special frequency region and it does not appear with the same reaction to all components of the spatial frequency 8 .Therefore, efficient encoding can be performed by the less sensitive area is set to Non-ROI region and the sensitive area is set to ROI region.In other words, we will use the characteristic which human eyes are not easy recognize the brightness change in the too dark and bright area.Increment threshold curve of Weber's law is shown Figure 3.

Setting the ROI and Non-ROI Area
Conventionally, it was used the form information or the image pattern for setting the ROI 9 .However, it was not considered at all for the 8X8 block which is used basic compression was adjusted within 2 dB to the Non-ROI PSNR.Seven video images were used to experiment.These were selected according to the complexity in the JPEG standard.The dispersion value of the current block can be obtained by equation ( 1).
And then, the dispersion of the most complex block can be calculated by equation ( 2).
Therefore, we can obtain the complexity R B .
Table 1 shows the spatial complexity values of 15 images which used the simulation.
Figure 5(a) is restored to a JPEG compression scheme the whole, and (b) is restored image which compressed in the proposed method.Comparing the two images, it can be seen that the resolution is good appears in ROI as a mouth, nose, eyes etc. Figure 7 shows that compare ROI region PSNR of the proposed compression method with the conventional JPEG.unit in DCT and quantization.So, 8X8 block standard deviation of the contrast was determined at setting the area by Weber's law.The smaller areas than the average of whole block standard deviation was setting to Non-ROI area.
Figure 4 shows ROI and Non-ROI area which are setting with a standard deviation of 8X8 block.

Experiments
The ROI area is very important in this simulation, but if it has a large compression difference between the ROI and Non-ROI area, many unexpected distortion can be occurred.So, the quantization size of ROI area

Conclusion
In this paper, we propose a new differential compression method using Weber's law.The proposed method is possible to maintain a high quality by differential compression of images.The image is separated simple and complex block using 8x8 block standard deviation.And also, we have applied low compression rate and high compression rate to each region.The proposed method is a simple calculation compared to conventional method which used various pattern of image block.In particular, we can apply to MPEG and H.26X because it is based on DCT technique.

Figure 5 (
b) is a little clearer than (a).

Figure 6 (
a) is compressed and restored face region by the JPEG system.And (b) is face region by the proposed method.

Figure 2 .
Figure 2. Flow chart of the proposed compression method.

Figure 4 .
Figure 4. ROI and Non-ROI with a standard deviation of 8X8 block.

Figure 5 .
Figure 5.Comparison of whole area.

Figure 6 .
Figure 6.Comparison of face area.

Table 1 .
Complexity of standard images