Truncated subband coding of images

• US 4,817,182 A
• Filed: 05/04/1987
• Issued: 03/28/1989
• Est. Priority Date: 05/04/1987
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Simoncelli, Eero P. M., Adelson, Edward H.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Couso, Jose L.
• Application Number:
  US07/045,987
• Time in Patent Office: 694 Days
• Field of Search: 382/31, 382/41-43, 382/49, 382/54, 382/56, 381/29, 381/31, 358/31, 358/38, 364/723-725, 364/728
• US Class Current: 382/248
• CPC Class Codes: H04N 1/646 Transmitting or storing colour television type signals, e.g. PAL, Lab; Their conversion into additive or subtractive colour signals or vice versa therefor H04N1/642, H04N1/644 take precedence
• CPC Class Codes: H04N 19/186 the unit being a colour or a chrominance component
• CPC Class Codes: H04N 19/635 characterised by filter definition or implementation details

Abstract

Image data is analyzed in a number of iterated analysis procedures, using two-dimensional quadrature mirror filters to separate low-pass spatial filter response component and three differently oriented high-pass spatial filter response components, which filter response components are decimated in both dimensions. The high-pass filter response components are coded as is. The low-pass filter response component is coded as is only in the last iteration; in the earlier analysis procedures the low-pass filter response component provides the input data for the secceeding analysis procedure.

Claims

1. A method for coding images consisting of sampled data concerning the intensities of a first closely packed two-dimensional array of similar-size picture elements, said method comprising a plurality p in number of successive analytic procedures each consisting of a respective combination of steps, which analytic procedures are consecutively ordinally numbered first through p^th in order of their performance, each of which analytic procedures includes in its respective combination of steps the steps of:

convolving a closely packed array of similar-size picture elements, which array is ordinally numbered the same as the analytic procedure this step is included in, with two-dimensional quadrature mirror filter kernels in order to decompose this array into a low-pass filter response component and three high-pass filter response components for the analytic procedure;

as a substep of said step of convolving, two-dimensionally decimating the three high-pass filter response components; and

encoding the three high-pass filter response components without any spatial-frequency analyzing substeps

each of which analytic procedures except the p^th includes in its step of convolving the substep of;

two dimensionally decimating the low-pass filter response component obtained during that analytic procedure, thereby to generate a closely packed array of similar-sized picture elements, which array is ordinally numbered the same as the succeeding analytic procedure and has picture elements larger than those in the array numbered the same as the analytic procedure this step is included in

and which p^th analytic procedure includes in its combination the step of;

encoding the low-pass filter response component obtained during the p^th analytic procedure.

2. A method as set forth in claim 1 wherein the step of encoding the low-pass filter response component obtained during the p^th analytic procedure includes the initial substep of two-dimensionally decimating that response.

3. A method as set forth in claim 2 wherein each said step of two-dimensionally decimating and said substep of two-dimensionally decimating all in effect subsample 2:

1 in one dimension of image space and subsequently subsample 2;

1 in the other, orthogonal dimension of two-dimensional image space.

4. A method as set forth in claim 1 wherein each said step of two-dimensionally decimating in effect subsamples 2:

1 in one dimension of image space and subsequently subsamples 2;

1 in the other, orthogonal dimension of two-dimensional image space.

5. A method for synthesizing a signal descriptive of images, from codes of the type generated by the method of claim 1, comprising a plurality p in number of consecutively ordinally numbered synthesis procedures performed in reverse order of their numbering, each of which synthesis procedures consists of a respective combination of steps, each of which analytic procedures includes in its respective combination of steps the steps of:

decoding respective ones of the codes to recover the three decimated high-pass filter response components for that synthesis procedure;

obtaining the decimated low-pass filter response component for that synthesis procedure;

interspersing the filter responses for that synthesis procedure among null samples and

convolving the resulting expanded sample matrices with two-dimensional quadrature mirror filter kernels, thereby to perform two-dimensional interpolations, and then

summing the results of these two-dimensional interpolations to generate a sum result, said sum result for each synthesis procedure with ordinal numbering higher than one being the decimated low-pass filter response for the synthesis procedure with next lowest ordinal number and being the synthesized signal descriptive of images for the first synthesis procedure

and said p^th synthesis procedure obtainings its decimated low-pass filter response by decoding the code encoding of the low-pass filter response component obtained during the p^th analysis procedure.

6. A method for coding polychromatic images consisting of sampled data converning the intensities in each of three color variables of a first closely packed two dimensional array of similar size picture elements, said method comprising the procedures of:

combining data concerning the three color variables for each picture element in a color matrixing procedure to generate a description of each picture element in terms of a luminance variable and orthogonal first and second chrominance variables;

analyzing said luminance variable in a plurality p in number of successive luminance variable analyzing subprocedures, consecutively ordinally numbered first through p^th in order of their performance, thereby to generate three high pass filter response components and a low pass filter response component for each luminance variable analyzing subprocedure;

two dimensionally low pass spatially filtering said first chrominance variable in a plurality q in number of successive first chrominance variable filtering subprocedures, which first chrominance variable filtering subprocedures are consecutively ordinally numbered first through q^th in order of their performance, q being a plural integer smaller than p by at least two;

analyzing said low pass spatially filtered first chrominance variable in a plurality (p-q) of successive first chrominance variable analyzing subprocedures consecutively ordinally numbered (q+1)^th through p^th in order of their performance, thereby to generate three high pass filter response components and a low pass filter response component for each said first chrominance variable analyzing subprocedure;

two-dimensionally low pass spatially filtering said second chrominanace variable in a plurality r in number of successive second chrominance variable filtering subprocedures, which second chrominance variable subprocedures are consecutively ordinally numbered first through r^th, r being a plural integer smaller than p by at least two;

analyzing said low pass spatially filtered second chrominance variable in a plurality (p-r) of successive second chrominance variable analyzing subprocedures consecutively ordinally numbered (r+1)^th through p^th in order of their performance, thereby to generate three high pass filter response components and a low pass filter response component for each said second chrominance variable analyzing subprocedure;

encoding the three high pass filter response components for each of said luminance variable, first chrominance variable and second chrominance variable analyzing subprocedures; and

encoding the respective low pass filter response components of said p^th luminance variable analyzing subprocedure, said p^th first chrominance variable analyzing subprocedure and said r^th second chrominance variable analyzing subprocedure

wherein each of said filtering subprocedures includes therein the respective steps of;

convolving a closely packed array of similar size picture elements, which array is ordinally numbered the same as the filtering subprocedure this step is included in with a two dimensional low pass filter kernel to obtain a low pass filter response; and

two dimensionally decimating the low pass filter response obtained during that subprocedure, thereby to generate a closely packed array of similar sized picture elements, which array is ordinally numbered consecutively after the ordinal number of that filtering subprocedure and has picture elements larger than those in the array ordinally numbered the same as that filtering subprocedure

wherein each of said analyzing subprocedures includes therein the respective steps of;

convolving a closely packed array of similar size picture elements, which array is ordinally numbered the same as the analytic subprocedure this step is included in, with two dimensional quadrature mirror filter kernels in order to decompose this array into a low pass filter response component and three high pass filter response components for the analytic subprocedure;

two-dimensionally decimating the three high pass filter response components obtained during that analytic subprocedure before said encoding of them; and

wherein each of which analytic subprocedures except the p^th includes the respective step of;

two-dimensionally decimating the low pass filter response component obtained during that analytic subprocedure, thereby to generate the one of said closely packed array of similar sized picture elements that is ordinally numbered the same as the succeeding analytic subprocedure and has picture elements larger than those in the array numbered the same as the analytic subprocedure this step is included in --and wherein each of said p^th analytic subprocedures includes the respective step of two dimensionally decimating the low pass filter response component obtained during that p^th analytic subprocedure before said encoding of it.

7. A method for coding polychromatic images as set forth in claim 6 wherein q and r are equal to each other.

8. A method for synthesizing signals descriptive of images, from codes of the type generated by the method of claim 6, comprising the procedures of:

synthesizing a luminance signal from the codes descriptive of said luminance variable in a plurality p in number of successive luminance-signal synthesizing subprocedures consecutively ordinally numbered first through p^th in reverse order of their performance;

synthesizing a first chrominance signal from the codes descriptive of said first chrominance variable in a plurality p in number of successive first-chrominance-signal synthesizing subprocedurres consecutively ordinally numbered first through q^th through p^th in reverse order of their performance of being an integer between one and (p-1); and

synthesizing a second chrominance signal from the codes descriptive of said second chrominance variable in a plurality p in number of successive second-chrominance-signal synthesizing subprocedures consecutively ordinally numbered first through r^th through p^th in reverse order of their performance, r being an integer between one and (p-1)

wherein each of said first through p^th luminance-signal synthesizing subprocedures, each of the (q+1)^th through p^th first-chrominance-signal synthesizing subprocedures, and each of the (r+1)^th through p^th second-chrominance-signal synthesizing subprocedures includes in its respective combination of steps the steps of;

decoding respective ones of the codes to recover the three decimated high-pass filter response components for that synthesizing subprocedure;

obtaining the decimated low-pass filter response component for that synthesizing subprocedure;

interspersing the filter responses components for that synthesizing subprocedure among null samples and

convolving the resulting expanded sample matrices with two-dimensional quadrature mirror filter kernels, thereby to perform two-dimensional interpolations, and then

summing the results of these two-dimensional interpolations to generate that synthesizing subprocedure result

wherein each of the first through q^th first-chrominance-signal synthesizing subprocedures and each of the first through r^th second-chrominance-signal synthesizing subprocedures includes in its respective combination of steps the steps of;

obtaining the decimated low-pass filter response component for that synthesizing subprocedure;

interspersing that low-pass filter response component among null samples; and

convolving the resulting expanded sample matrix with a low-pass filter kernel, thereby to perform a two-dimensional interpolation that generates that synthesizing subprocedure result

wherein the p^th synthesizing subprocedure in each of said synthesizing procedures the respective decimated low-pass filter response is obtained by decoding the code encoding the low-pass filter response component obtained during the corresponding p^th analyzing subprocedure

wherein, within each of said synthesizing subprocedures except the first in a synthesis procedure, the synthesizing subprocedure result is used as the decimated low-pass filter response component for the synthesizing subprocedure with next lower ordinal numbering

and wherein the first synthesizing subprocedure results of said procedures of synthesizing a luminance signal, of synthesizing a first chrominance signal, and of synthesizing a second chrominance signal are these signals respectively.

9. A method for synthesizing signals descriptive of images as set forth in claim 8 wherein q and r are equal to each other.