Image analysis – Image compression or coding
Reexamination Certificate
1999-11-10
2004-11-02
Couso, Yon J. (Department: 2625)
Image analysis
Image compression or coding
C382S240000, C382S305000, C707S793000
Reexamination Certificate
active
06813384
ABSTRACT:
BACKGROUND
This invention relates generally to image processing and computer graphics. More specifically, the invention relates to compression of video using wavelet transforms.
In general, video compression techniques reduce the bit rate used for storing or transmitting video while hopefully maintaining acceptable video quality. Compression of video normally involves transforming the video data to reduce correlation between adjacent pixels. The transformed data may then be quantized to reduce the amount of data. The quantized data is then encoded for transmission or storage.
In a discrete cosine transform (DCT) a plurality of small blocks are each subjected to a transformation. In discrete wavelet transformation (DWT) the full and entire frame may be transformed. DCT differs from DWT in that only in DWT can one develop the image from the transformed data.
DWT is a multi-level, multi-resolution decomposition technique. Each row of a video image frame is subjected to low pass filtering followed by the elimination of alternating coefficients. Each row is then subjected to high pass filtering followed by the elimination of alternating coefficients. Each row of the image frame then has a set of low pass filter coefficients followed by a set of high pass filter coefficients. The low pass and high pass filtering are then repeated in the vertical dimension with alternating coefficients eliminated as described above.
A result of the low pass filtering with down-sampling in the horizontal dimension is a horizontally narrowed image having the same height as the original image. A result of low pass filtering with down-sampling in the vertical dimension of the horizontal low pass filtered image, a scaled image may be created which is reduced in both height and width. However, the image may otherwise be a reasonably accurate scale depiction of the original image. High pass filtering creates edge detail.
Thus DWT may produce four sub-bands including the LL sub-band which is the scaled version of the original image and results from one dimensional low pass filtering in both the horizontal and vertical dimensions. The LH sub-band, containing vertical edge information, is the result of low pass filtering in the horizontal dimension and high pass filtering in the vertical dimension. The HL sub-band, containing horizontal edge information, is the result of high pass filtering in the horizontal direction and low pass filtering in the vertical direction. The HH sub-band is mostly high frequency noise.
DWT may be repeatedly applied to each newly created LL sub-band, as one example. Thus a level one DWT decomposition may include the four sub-bands LL, HL, LH and HH. A level two decomposition may include seven sub-bands and a level three decomposition may include ten sub-bands and so on.
After transformation, the video data may be subjected to quantization and entropy or “variable length” encoding. In variable length encoding, shorter code lengths are allocated to the most frequently utilized symbols. Longer code lengths are allocated to less frequently used symbols. Thus, as a result, the encoded signal portions have lengths which may be variable throughout the data stream depending on the frequency with which particular symbols happen to be utilized.
In a number of cases using DWT, it is desirable to access a particular sub-band. However because of variable length encoding, each compressed frame and each compressed video sequence, can only be sequentially processed. That is, each compressed frame is not accessible before all of its previous frames have been decompressed.
This makes video processing more difficult. For example, non-sequential frame decompression, fast forward and reverse, low resolution preview, display frame buffer control, frame dropping or skipping, non-linear video processing, editing and retrieving are more difficult and time-consuming because it is necessary to decompress all the previous frames before any information can be reviewed. This limits the efficiency of the compressed video handling.
Particularly in connection with DWT, it is often desirable to access the LL-band since the LL-band provides a scaled version of the original image. However again, because of variable length encoding, after the compression is done, the only way to access any particular piece of information within the sequence is to decompress the entire sequence.
Thus there is a continuing need to enable the location information in discrete wavelet transformation compressed data without necessitating the decompression of each sequential frame.
SUMMARY
In accordance with one aspect, a method includes computing size information for a discrete wavelet transform base compressed data frame. The compressed data frame is indexed with the size information.
Other aspects are set forth in the accompanying detailed description and claims.
REFERENCES:
patent: 4751583 (1988-06-01), Levine
patent: 5321776 (1994-06-01), Shapiro
patent: 5915038 (1999-06-01), Abdel-Mottaleb et al.
Acharya Tinku
Tan Yap-Peng
Tsai Ping-Sing
LandOfFree
Indexing wavelet compressed video for efficient data handling does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Indexing wavelet compressed video for efficient data handling, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Indexing wavelet compressed video for efficient data handling will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3351269