Image analysis – Image compression or coding – Pyramid – hierarchy – or tree structure
Reexamination Certificate
2000-04-24
2004-01-20
Do, Anh Hong (Department: 2624)
Image analysis
Image compression or coding
Pyramid, hierarchy, or tree structure
C382S232000, C382S248000
Reexamination Certificate
active
06681051
ABSTRACT:
BACKGROUND OF THE INVENTION
In digital photography of single pictures, i.e. motionless still pictures, there is the problem that the required data quantities are very large. This particularly applies when the resolution in digital photography is to approximate that of chemical photography. A known standard for data compression of still pictures is the JPEG standard. This standard does not achieve sufficiently large data reduction rates and, moreover, has the drawback that a data reduction in fixed picture blocks of, for example, 8×8 pixels is performed, which in the case of strong compression can be recognized as artefacts in the reconstructed picture.
To achieve higher compression rates, the Discrete Wavelet Transformation, referred hereinafter as DWT, is known. It has proved to be an efficient method in picture coding and compression. In wavelet transformation of the picture data, so-called wavelet coefficients or sub-bands are generated which can be subsequently quantized and entropy-encoded. The resultant compressed data stream may be used for transmission or storage. The DWT is the basis for future compression standards such as, for example, JPEG 2000 and is an alternative to the known Discrete Cosine Transform (DCT). In picture processing, the so-called two-dimensional Discrete Wavelet Transformation, hereinafter referred to as 2-D DWT, is used. This may be understood to be a consecutive application of a one-dimensional wavelet transformation in the horizontal and vertical directions.
The following aspects are essentially known from the state of the art.
On the one hand, the 2-D DWT can be performed by way of a complete, twice consecutive application of a simple DWT in the horizontal and vertical directions. However, the result is the necessity of buffering the data of the sub-bands between the first and the second transformation stage; thus a considerable memory volume is required. Furthermore, the bus load to and from which the data are copied is increased by at least a factor of 2.
Moreover, it is possible to perform a direct 2-D DWT in which the transformation is performed in one step. It is true that the bus load is reduced thereby but this poses the problem that a comparatively large part of the picture must be simultaneously available for transformation, so that it must be internally stored when it is being processed. In the state of the art, line memories or other memories are used for this purpose. This involves the problem that the size of the internal buffer memories borders on the maximum picture resolution to be processed and that these buffer memories are required as additional memories.
When the transformation is performed in a multiple, iterative way, in which the single sub-bands determined in a transformation operation are buffered time and again, the required memory volume for the initial picture is increased by approximately one-fourth. The sub-pictures are then rewritten into the picture memory. The picture memory must then, however, be implemented for the higher bit depth. Since only picture data, which are no longer necessary for further transformations, can be overwritten when the sub-pictures are being rewritten into the picture memory, the sub-pictures can be stored in the memory in a strongly fragmented form only. A possible subsequent re-assortment considerably increases the bus load, the memory space required for buffering and the computation time.
SUMMARY OF THE INVENTION
It is an object of the invention to provide an arrangement for transforming picture data which is suitable for the two-dimensional Discrete Wavelet Transformation and generates a minimum load of the memory bus, minimizing the overall required memory space, while the resolution of the picture to be processed is not limited by internal register sizes and its architecture provides a transformation without any losses.
According to the invention, this object is solved in that a picture memory is provided in which the data of a picture are stored prior to the start of the first transformation plane and in which the data of a further sub-band to be transformed are stored after the first transformation plane, in which process the data of the picture are partially overwritten, in that the data of a possible further sub-band to be segmented are stored in the picture memory after every further transformation plane, in that a sub-band memory is provided in which, after a transformation process of a plane, the data gained during this transformation of those sub-bands which are no longer to be segmented in further transformation planes are stored, which data of said sub-band are stored adjacent to each other and in which sub-band data determined in previous transformation planes and possibly already stored in the sub-band memory are not overwritten, in that, in the transformation planes, the data of the picture or the sub-band data stored in the last transformation plane are read from the picture memory, which data are read in blocks comprising a basic block having a size corresponding to the picture section or sub-band to be transformed, and a frame surrounding said picture section and having a width corresponding to half the maximum filter depth of the filters used for the transformation, and in that all basic blocks combined cover all pixels of the picture or sub-band data in the picture memory.
In the arrangement according to the invention, two memories are provided, namely one picture memory and one sub-band memory. The picture memory is provided to take up picture data of the picture to be transformed before the start of the first transformation plane. In every subsequent transformation plane, the sub-band to be further transformed is written into this picture memory. Since the original picture and the sub-band data stored in the picture memory during the previous transformation planes are no longer required for the subsequent transformation planes, the data of the picture or the data of the sub-band of the previous transformation plane can be overwritten. Thus, this picture memory can be dimensioned in such a way that, as far as its size is concerned, it is adequately dimensioned for taking up the original picture data. Consequently, no additional storage quantity is required in the picture memory for those data, to be stored in each transformation plane, of that sub-band which is to be further transformed.
After each transformation process of a plane, the data, gained during this transformation, of those sub-bands which are not to be subjected to a further transformation are stored in the sub-band memory. The data can then be stored adjacent to each other and can be stored in an ordered way or in the desired way in the sub-band memory so that no reassortment is required prior to reading the sub-band data.
When the transformations are being performed in the relevant transformation planes, not all data of the picture or of the sub-band stored in the picture memory and to be further transformed are read from the picture memory, but these data are read in blocks only and the transformation is performed for data of these blocks. However, to ensure that this block structure does not influence the transformation process, i.e. no block structure or similar disturbances appear in the reconstructed picture, the blocks are formed in such a way that they comprise a basic block having a size corresponding to the part of the picture or the sub-band to be transformed. This basic block thus comprises that part of the picture or sub-band which is to be transformed. Additionally, this basic block is surrounded by a frame which comprises so many pixels towards all sides that it has the maximum half filtering depth of the filters used in the transformation process. It is thereby ensured that the transformation process which is applied to the entire block is performed in such a way that the transformation can be performed for the data of the basic block without any disturbing effects by the block. The transformation for the pixels in the basic blocks is thus not influenced at all by the block
Götting Detlef
Hoppe Dirk
Biren Steven R.
Do Anh Hong
LandOfFree
Arrangement for transforming picture 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 Arrangement for transforming picture data, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Arrangement for transforming picture data will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3205190