Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
  • H04N19/59—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial sub-sampling or interpolation, e.g. alteration of picture size or resolution
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
  • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
  • H04N19/132—Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
  • H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
  • H04N19/635—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets characterised by filter definition or implementation details

Definitions

• the invention generally relates to method and device for a pyramidal image coding aimed at application in systems for compression of digitized photo and TV images when they are transmitted via digital communication nets: ISDN, ATM, Internet, terrestrial, satellite and cable communication channels, for nonlinear editing systems, control and security systems, videotelephone and videoconference systems, archivating and editing of digital images on hard disks, CDs, magnetic tapes, etc.
• the first contains information about the low-frequency components in the horizontal direction of the spectrum and about the high- frequency components - in the vertical direction; the second one - about the high-frequency components in the horizontal direction and about the low-frequency ones - in the vertical direction; the third - about the high-frequency components in the diagonal direction; and the fourth - about the low-frequency components in horizontal and vertical directions simultaneously.
• the image that corresponds with the last band is used as an input (base) for the recurrent composition of the next pyramid level and at the end of the decomposition it is the top of the pyramid.
• the image compression is attained when the quantizated pixels of the Wavelet pyramid levels are transmitted via the communication channel. When the decompression is performed these pixels are dequantizated and processed in inverse order by special circuits for supersampling and by a bank of quadrature-mirror filters (QMF). In result is obtained a recursive image restoration.
• QMF quadrature-mirror filters
• a device [4] that performs the above described method whose input is connected in parallel with two digital filters for image processing in horizontal direction: the first filter is a high-frequency and the second - a low-frequency one.
• the outputs of the two filters are connected through circuits for double decimation, respectively with two couples of digital filters connected in parallel (a high-frequency and a low - frquency one) for image processing in the vertical direction.
• the four outputs of these two couples of filters are connected through circuits for double decimation with quantization circuits whose outputs are connected through a multiplexer with the communication channel.
• the output of the fourth quantization circuit is connected by a circuit for inverse quantization and a buffer videomemory with the coder input.
• the input of the decoder is connected through a demultiplexer with four circuits for inverse quantization, double supersampling (interpolation) and filtration by a bank of inverse quadrature- mirror filters. Four of them are used for image processing in the horizontal direction and two - for image processing in the vertical direction but the action of the second group of filters begins only after the end of the second interpolation of the images, already filtered in horizontal direction by the first group of filters.
• the input image is accepted as zero level (basis) of the auxiliary
• Gaussian pyramid In the process of the pyramidal coding the image size is reduced iteratively through successively applying of two-dimentional low-frequency filtration and decimation. In result of the consecutive filtrations and decimations every second pixel in horizontal and vertical directions is removed and in result the spatial image resolution is reduced twice. After the first reduction the image corresponds to the first level
• Every consecutive pyramid level is obtained from the lower (preceding) one by reduction of the size of the corresponding image with low-frequency filtration and decimation. The reduction is stopped when after a consecutive performance the image size becomes equal to the minimuin possible size of one pixel (i. e. its size is lxl pixels).
• This image is the last level of the Gaussian pyramid and corresponds to its top. Together with the building of the
• Gaussian pyramid is constituted the so called "Laplacian" image pyramid.
• Its zero level (base) is the difference between the images of the zero level of the Gaussian pyramid and the next (first) level, expanded twice in horizontal and vertical direction by applying of two-dimentional interpolation.
• the next (first) level of the Laplacian pyramid is obtained in the same way as the zero level and is the difference between the first level of the Gaussian pyramid and its following (second) level, expanded twice in result of the applied two-dimentional interpolation, etc.
• the last level (the top) of the pyramid is an image of size (lxl) pixels that coincides with that of the Gaussian one.
• the compression of the image decomposed with Laplacian pyramid is obtained with the reduction of 1/4 of the total number of the pyramid pixels (one of every 4 pixels in each level is removed) in correspondence with an algorithm called “Reduced Difference Pyramid” (RDP) [2].
• RDP Reduced Difference Pyramid
• For the further pyramid compression are used the operations “Quantization” of the elements of every level, and “Entropy coding” [6].
• the data from the compressed pyramid levels is transmitted via the communication channel consecutively from its top to the base and in this way is performed the so called “progressive" image transmitting with gradually increasing of its resolution and size.
• the received data are processed with the operations "Inverse quantization” and "Entropy decoding".
• the first received image that corresponds to the top of the pyramid (its last level) is expanded twice by two-dimentional interpolation and is added to the next image from the last but one level of the Laplacian pyramid, etc. In this way consecutively are restored the images that correspond to the levels of the Gaussian pyramid until its base is reached (the zero level).
• a device that performs the above described method and contains an image source connected with the inpuit of a coder for recursive calculation of the Gaussian and Laplacian pyramids levels.
• the coder itself contains consecutively connected circuits: for image reduction, quantization, inverse quantization and expansion.
• the output of the commutation circuit is connected with the adding input of a summator whose subtracting input is connected with the output of the expanding circuit.
• the output of this summator is connected with a quantization circuit whose output is the output of the coder connected with the channel used for transmission of the difference images to the decoder.
• the output of the quantization circuit connected with the reducing circuit is also connected through a buffer memory with the input of the coder.
• the decoder used for the recursive restoration of the transmitted image whose input is connected with the communication channel contains an input circuit fot inverse quantization. It is connected with the first input of a summator whose second input is connected with the output of an expanding circuit. Its input is connected with the output of a buffer memory whose input is connected with the decoder output.
• the essence of the invention is a method for pyramidal image coding, based on new kind of digital image decomposition, called Inverse differential pyramid.
• the digital input image is approximated by one of the two following ways: with a two-dimentional polynomial function of m coefficients whose values are determined from the requirement for minimum mean-square error, or with an image obtained after inverse orthogonal transform of the filtrated transform (the discrete spectrum) of the input image after retaining a group of only m selected low-frequency coefficients in case that this transform can be determined with every one of the known linear orthogonal transforms as for instance are the Discrete cosine transform (DCT), the Walsh-Hadamard transform (WHY), etc.
• DCT Discrete cosine transform
• WHY Walsh-Hadamard transform
• the approximating image is described with m coefficients regardless of the selected kind of approximation. These coefficients represent the "zero" level (the top) of the inverse differential pyramid.
• the approximation image obtained from the coefficients of the "zero” level is subtracted element by element from its corresponding source input image.
• the resulting "zero" difference image is divided in four subimages of same size and form. Every subimage is processed in the same way as the input image in the previous stage. In result is obtained the "zero" level difference that is described with Am coefficients. These last coefficients determine the first pyramid level.
• the second pyramid level is obtained when the approximation of the "zero" difference is subtracted by the "zero” difference itself and thus obtained "first" difference image is divided in 16 subimages with equal size and fo ⁇ n. Each subimage is processed with the same approximation operations as the preceding "zero” level, etc. If the input image has 2"x2 n pixels, the last but one (n-l)-th level of the corresponding inverse pyramid has a corresponding difference image n-2 that is described with 4 n" 'm coefficients. The last pyramid level is the residual difference image n-1 whose size is equal with that of the input one.
• the pyramid built this way has n+1 levels and is described with 4 n+1 (m/3) coefficients that are enough for the full restoration (lossless) of image of any kind.
• n+1 (m/3) coefficients that are enough for the full restoration (lossless) of image of any kind.
• 1/16 of the coefficients of the neighbour subimages are reduced on the basis of their correlation and the remaining coefficients are processed with entropy coding.
• the so called visually lossless coding In order to get higher compression is used the so called visually lossless coding.
• Fot this purpose is used bi-directional truncation of the low-information levels from the pyramid top and base, and quantization of the coefficients in the remaining levels before their entropy coding.
• the data from the pyramid levels is transmitted via the communication channel in direction from its top to the base.
• the image decoding is performed over received data and it consists of consecutive entropy coding and inverse quantization (the last is used only in case of visually lossless coding).
• coefficients of the subimages from the transmitted pyramid levels are restored their corresponding polynomial functions or approximating images, calculated on the basis of the inverse orthogonal transform of their coefficients.
• the restored images for every pyramid level are added consecutively in direction from top to the base of the pyramid and in result is obtained the source image reproduced with gradually increasing resolution corresponding to the number of the transmitted pyramid level.
• the kind of the subimage approximation - polynomial or orthogonal transform - must be dete ⁇ nined in advance in accordance with the requirements for the time required for the compression and with the approximation precision.
• the polynomial approximation ensures higher accuracy that results in restored image with better quality but requires more calculations.
• the approximation on the basis of orthogonal transform is preferred when the application requires higher compression speed (for real-time applications) and some image quality deterioration is acceptable.
• the method facilitates the progressive image transmitting because the pyramid levels are dete ⁇ nined consecutively from its top to the base in inverse order compared with the Laplacian pyramid;
• the pyramid levels contain coefficients of transforms or polynomial functions whose changes (due to random errors got in result of their transmission via the communication channel) have small influence over the visual quality of the image compared with the errors in the Laplacian pyramid because its levels are difference images.
• the method ensures higher compression coefficient than the Laplacian pyramid and than the "Reduced difference pyramid” [2] for same accuracy (quality) of the restored image.
• the essence of the invention is in a device for pyramidal coding too, that contains an image source, connected with the input of a coder for recursive calculation of the inverse differential pyramid.
• the coder consists of first commutatoi whose first input is connected with the image source and whose output is connected with the input of the first videomemory and with the first input of the second commutator.
• the output of the first videomemory is connected with the second inputs of the second commutator and of the first summator.
• the output of the second commutator through an electronic switch is connected with the first input of a circuit for calculation of the polynomial coefficients or the coefficients of a "truncated" orthogonal transform.
• the second input of the same circuit is connected with the output of the first circuit for definition of weighted or
• the output of the last circuit is connected with the first input of the third commutator and with the input of a circuit for inverse quantization whose output is connected with the input of the second videomemory.
• the output of this memory is connected with the first input of the first circuit for polynomial restoration or inverse orthogonal transform whose second input is connected with the output of the first circuit for definition of weighted or "base” images and its output is connected with the first subtracting input of the first summator.
• This output is connected correspondingly with the second inputs of the first and tliird commutators.
• the output of the third commutator is connected through the circuit for entropy coding with the output of the coder for the recursive calculation of the inverse differential pyramid.
• This output is connected through communication channel with the corresponding pyramidal decoder.
• This decoder consists of circuit for entropy coding whose input is the input of the pyramidal decoder and whose output is connected with the input of a demultiplexer.
• first output is connected with the input of a second circuit for inverse quantization whose output is connected through third videomemory with the first input of the second circuit for polynomial restoration or inverse orthogonal transform, and whose second input is connected with the output of a second circuit for definition of weighted or "base" images.
• the output of the second circuit for polynomial restoration or inverse orthogonal transform is connected with the first input of fourth commutator whose second input is connected with the second output of the demultiplexer and its output is connected with the first input of a second summator whose second input is connected with the output of the fourth videomemory. Its input is connected with the output of the second summator and is the output of the pyramidal decoder.
• the advantage of the device that performs the method is the more simple structure compared with that of a similar device for pyramidal image coding with Laplacian pyramid and especially in the case of including of closed loop for the quantization noise [7].
• This advantage is due to the canonical structure of the inverse differential pyramid and the lack of circuits for image reduction and expansion used in the Laplacian pyramid. These circuits are performed with decimators, interpolators and two-dimentional digital filters with minimum size of 5x5 elements that require large number of computations.
• Fig. 1 is the block diagram of the device for pyramidal image coding based on decomposition called inverse differential pyramid for progressive image transmitting.
• Fig. 1 is a block diagram of the device consisting of coder and decoder. They include four two-input commutators 1 , four videomemories 2, a circuit for definition of the polynomial coefficients or of these of the "truncated" orthogonal transform 3, two circuits for polynomial reconstruction or inverse orthogonal transform 4, circuits for quantization and inverse quantization correspondingly 5 and 6, two circuits for definition of weighted or "base” images 7, two summators 8, circuits for entropy coding and decoding correspondingly 9 and 10, demultiplexer
• the image source is connected with the first input a of the first commutator 1 and its output b is connected with the input of the first videomemory 2 and with the first input of the second commutator 1.
• the output c of the first videomemory 2 is connected correspondingly with the second inputs of the second commutator 1 and of the first summator 8.
• the output of the secont commutator d through electronic switch 12 is connected with the first input e of the circuit for definition of the polynomial coefficients or the coefficients of the "tRyered" orthogonal transform 3.
• the second input of the same circuit 3 is connected with the output m of the first circuit for definition of weighted or "base” images 7 and its output /is connected with the input of the quantization circuit 5.
• the output g of the last circuit is connected with the first input of the third commutator 1 and with the input of the circuit for inverse quantization 6 whose output h is connected with the input of the second videomemory 2.
• Its output k is connected with the first input of the first circuit for polynomial reconstruction or inverse orthogonal transform 4 whose second input is connected with the output m of the first circuit for definition of weighted or "base” images 7, and its output / is connected with the first subtracting input of the first summator 8. Its output is connected with the second inputs correspondingly of the first and third commutators 1.
• the output n of the third commutator 1 is connected through a circuit for entropy coding 9 with the output p of the coder that performs the recursive calculation of the inverse differential pyramid. This output p is connected through the communication channel with the input of the corresponding pyramidal decoder.
• This decoder consists of circuit for entropy coding 10 whose input is the input of the pyramidal decoder and whose output x is connected with the input of the demultiplexer 11. Its first output r is connected with the input of the second circuit for inverse quantization 6 whose output s is connected through third videomemory 2 with the first input t of the second circuit for polynomial restoration or inverse orthogonal transform 4, whose second input is connected with output m of the second circuit for defmitioin of weighted or "base" images 7.
• the output u of the second circuit for polynomial reconstruction or inverse orthogonal transform 4 is connected with the first input of fourth commutator 1 whose second input q is connected with the second output of the demultiplexer 1 1 and its output v is connected with the first input of the second summator 8 whose second input y is connected with the output of fourth videomemory 2. Its input w is connected with the output of the second summator 2 and is itself the output of the pyramidal decoder.
• the action of the device in accordance with the invention is as follows: In the first cycle of the recursive image processing the signal is applied to the input of the pyramidal coder through the input commutator 1 and is stored in the videomemory 2. At the same time the image signal passes through the second commutator 1 and the switch 12 and enters the circuit that calculates the polynomial coefficients or those of the "truncated" orthogonal transform 3. For this purpose at the input of the same circuit are applied simultaneously the weighted or "base" images obtained in the circuit 7.
• the kind of these images in the first type of approximation is defined in advance in accordance with the order of the polynomial regression function and with the input image (or subimage) size, and for the second type - in accordance with the chosen linear orthogonal transform (for example Discrete Cosine
• base image in accordance with the type of the chosen approximation - polynomial or with "truncated" orthogonal transfo ⁇ n.
• the coefficients number is dete ⁇ nined in advance and for the first type of approximation it depends on the selected polynomial function, for example - six coefficients for a surface of second order, three coefficients for a plane, etc.
• the number of coefficients that corresponds with the low-frequency spatial coefficients in the image (or subimage) transform is selectable, for example it can be four, three, etc. If the number of coefficients is higher the approximation is more exact but this decreases the compression coefficient.
• the coefficients number is always settled by compromise and for example it can be 4 for all pyramid levels.
• the values of these coefficients are calculated when the input image is already saved in the videomemory 2.
• the coefficients obtained at the output / of circuit 3 are quantizated in circuit 5.
• the coefficients from the output g of circuit 5 pass through the commutator 1 and enter the entropy coding circuit 9.
• the quantizated coefficients from output g are dequantizated in the circuit for inverse quantization 6 and are stored in the videomemory 2 used for saving the coefficients of the approximating image (or subimage).
• the two videomemories (the one, used for saving the image, and the other one, that stores the approximation coefficients) are read simultaneously and the coefficients from the output k of the second one are processed in the circuit for inverse orthogonal transform or for restoration of the polynomial functions 4 in order to get the corresponding approximating image (or subimage).
• the output m of the circuit 7 is sent a signal to the circuit 5, indicating the corresponding "base" image.
• the signal from the output / of the circuit 4 is subtracted in the summator 8 from that of the output c of the videomemory 2.
• the difference signal on the output i of the summator 8 passes through the input commutator 1 and is stored in the input videomemory 2 substituting the image information obtained in result of the preceding first cycle of the recursive calculation of the levels in the inverse differential pyramid.
• the input memory that contains the zero difference image must be read twice: at first the data is arranged in consecutive blocks that correspond to its four subimages.
• These images illustrate the method and permit the visual evaluation of the restoration quality on every stage of their progressive transmitting.
• the method and the device in accordance with the invention pe ⁇ nit the performance of pyramidal coding without losses (or without visual losses) for still images and for videosequences that can be greyscale or colour.
• the image is described with three components (matrices) every one of which can be processed in order to be compressed in accordance with the method and the device.
• the described method and the corresponding device are used for compression of videosequences (moving images) every single frame has to be processed intraframe and in real time.

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Compression Or Coding Systems Of Tv Signals (AREA)
• Image Processing (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Priority Applications (1)

Applications Claiming Priority (2)

Publications (1)

Family

ID=3927850

Family Applications (1)

Country Status (3)

Cited By (4)

Citations (1)

• 1999
  • 1999-07-29 BG BG103619A patent/BG63586B1/bg unknown
  • 1999-12-09 AU AU16454/00A patent/AU1645400A/en not_active Abandoned
  • 1999-12-09 WO PCT/BG1999/000027 patent/WO2001010130A1/en active Application Filing

Patent Citations (1)

Non-Patent Citations (6)

Also Published As

Similar Documents

Legal Events