Methods for quantizing and compressing digital image data

Details Methods for quantizing and compressing digital image data Methods for quantizing and compressing digital image data

2000-06-05

2002-05-28

Young, Brian (Department: 2819)

Coded data generation or conversion

Digital code to digital code converters

To or from differential codes

Reexamination Certificate

active

06396422

ABSTRACT:

FIELD OF THE INVENTION
The present invention relates to the quantization and compression of digital image data, and more specifically to image data containing inherent signal dependent noise and color digital image data.
BACKGROUND OF THE INVENTION
Digital images contain huge amounts of data, especially for high quality display and printing. Compression is useful to reduce the time for transferring data in networks and for saving digital media storage space. Images may be monochromatic or colored and colored images may have two or more channels. The most common, images are RGB images (for display) and CMYK images (for print). The common source of most images is an input device such as a scanner or digital camera. Such input devices are two-dimensional instances of the class of applications where signal is captured by a light-sensitive device.
In video capture, digital photography and digital scanning, the image is acquired by means of an electronic sensor such as a charge-coupled device (CCD) or a complimentary metal-oxide semiconductor (CMOS) device with cells sensitive to electromagnetic radiation such as visible light, infrared or ultra-violet waves. In medical or astronomical imaging, images may also be acquired by X-ray, microwave or radiowave sensitive detectors. Such images will have inherent characteristic noise that is not a part of the original signal reaching the sensor.
A typical noise characteristic, as found in particle-counting devices, such as a CCD sensor, is mainly a combination of ‘dark current’ shot noise and photon shot noise. CCD noise characteristics are described in Theuwissen, A. J. P, “Solid-State Imaging with Charge-Coupled Devices”, Kluwer (1995), sections 3.4 and 8.1, incorporated by reference herein. Dark current shot noise is constant for all light levels and is therefore significant at low light levels, whereas photon shot noise is proportional to the square root of the light level and is therefore most significant at high light levels.
In most electronic light sensing devices, such as video cameras for example, photon shot noise is largely eliminated from the output by use of a logarithmic or similar analog amplifier placed before the analog-to-digital (A-D) converter. Differences due to noise of the order of the square root of the incident signal level are thereby compressed into a nearly noiseless signal. Such non-linear compression of image signals causes little loss of visual quality because of the way the human eye interprets images. Specifically, in this cases the dynamic range of the original lighting is usually many orders of magnitude larger than the dynamic range of the projection or display device. Since the human eye is non-linear, in fact close to logarithmic, in its response to different dynamic ranges, the original signal would have to be non-lineally compressed for viewing or printing, after A-D conversion, to maintain subjective fidelity.
In modern digital cameras and scanners, such as the Leaf Volare and the EverSmart Supreme scanner, both devices available from Scitex Corporation, Herzlia, Israel, the CCD signals are converted to 14 or more bit data without non-linear, analog amplification. Thus the digitized data is directly proportional to the number of electrons captured by the sensing device. In this case there is inherent characteristic noise in the digital data. It is foreseeable that other devices such as, for example, medical or astronomical devices will also make the change to digitizing data without non-linear amplification. Many-bit files are large and require extensive storage space. It is therefore desirable to compress the data, provided there will be little or no loss of signal information.
The need to compress has also arisen for high-quality, multi-component images such as RGB images (for display) or CMYK images (for print), consisting of 8 (eight) bits per component. If the origin of the data is captured light intensities as in scanners or digital cameras, the data are transformations of the original light intensities, designed to allow display or printing of the image in a visually optimal manner. With the proliferation of such images and the increase in digital communication, a need has arisen to compress them for transmission and storage.
The most common means for compressing multi-component images is by use of the JPEG standard. The JPEG standard is described in international standard IS 10918-1 (ITU-T T.81), available from the Joint Photographic Experts Group, http.//www.jpeg.org.
The lossy JPEG standard is a method that compresses images by impairing high frequency data that are not noticeable to the eye. Improvements of lossy JPEG have also been suggested which utilize subtler psychovisual criteria. A summary is provided in T. D. Tran and R. Safranek, “A Locally Adaptive Perceptual Masking Threshold Model For Image Coding,” Proc. IEEE Int Conf. on Acoustics, Speech, and Signal Processing, Vol. IV, pp. 1882-1886, Atlanta, May 1996. These improvements are intra-component additions. No use is made of inter-component, i.e. color, information.
JPEG compression can optionally utilize a color psychovisual criterion in case of compression of an RGB image. A matrix transformation can be applied to the RGB value to yield a luminance component and two color components. Then the color components can be optionally sampled at half the rate of the luminance component in the subsequent compression processing and the quantization tables for luminance and color may be different. This is in accordance with the fact that the human eye perceives color at less resolution than luminance.
The described procedure is however not optimal in three senses. First, the matrix transformation does not map the RGB image to a psychovisually uniform space if the RGB data are proportional to actual light intensities. Only if the RGB are logarithmic or power transformations of light intensities, is the image of the mapping approximately psychovisual and then only in the luminance component. Second, the lower rate of sampling of the color components completely eliminates color modulation of more than half the original frequency. Thirdly, the transformation cannot be used for color spaces with more than three components, such as CMYK data.
In addition lossy JPEG does not have the property that once-compressed and twice-compressed images are identical. This is desirable for applications where the file needs to be compressed and decompressed several times. Also JPEG compression cannot guarantee a maximum deviation of a decompressed pixel from the original pixel.
As an alternative to JPEG, digital image data may be losslessly or nearly losslessly compressed by a variety of compression algorithms. Fast and proven effective methods, especially tuned to compress images effectively, are described in U.S. Pat. Nos. 5,680,129 (Weinberger, et al.) and U.S. Pat. No. 5,835,034 (Seroussi, et al.). The methods disclosed in these patents form the basis for the draft standard JPEG-LS of ISO, ISO/IEC JTC1/SC29 WG1 (JPEG/JBIG) “Information Technology—Lossless and Near-Lossless Compression of Continuous-Tone Still Images, Draft International Standard DIS14495-1 (JPEG-LS)”(1998) (hereinafter referred to as “Draft International Standard DIS14495-1”), incorporated by reference herein, and attached hereto as Addendum I.
Another lossless
ear-lossless method for images is described in Wu, “Lossless Compression of Continuous-Tone Images Via Context Selection, Quantization And Modeling”, IEEE Transactions on Image Processing, Vol. 6, No. 5, May 1997.
The above methods are both differential pulse code modulation (DPCM) techniques. DCPM is a general signal compression technique based on the formulation of predictors for each signal value from nearby values, The difference between the predicted value and the true value is encoded. DPCM techniques in general and for images in particular, are described in Jain, A. K., “Fundamentals of Image Processing”, Chapter 11, Prentice-Hall (1989), the entire work incorporated by reference herein.
DPCM techniques do not give good c

Affiliated with

Also associated with

Rating

Methods for quantizing and compressing digital image data does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Methods for quantizing and compressing digital image data, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods for quantizing and compressing digital image data will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2850669