Multiple module architecture for implementing full search block matching
AssigneeIndustrial Technology Research Institute
InventorsVincent M. S. Lin, Foo-Ming Lee
Life of this patent
Estimating the term…
Abstract
A circuit implementation of a block matching algorithm includes a plurality of modules. Each module is a one-dimensional systolic array. The modules can be connected in tandem to increase the power of computation so that the best match of a current block in a search window can be obtained more rapidly without increasing the number of input ports. Further, the modules can be connected in tandem so as to enable the size of the search window to be increased.
Claims24
A full search block matching circuit comprising:
- a plurality of m(m>1) modules connected in tandem, each of said modules comprising a one-dimensional array of processing elements,
- a first input for inputting pixel values from a block of pixels in a current video frame in a predetermined sequence to said modules,
- a second input for inputting pixel values from a search window of a previous video frame in a predetermined sequence to said modules,
- a control circuit for generating a stream of mask signals for each of said modules, and
- at least one gate circuit for receiving said pixels from said first and second inputs and said mask signals, and for selectively transmitting particular pixels inputted at said inputs to particular ones of said modules such that each one of said modules being enabled for equal, staggered periods of time.
A full search block matching circuit comprising:
- a plurality of m(m>1) modules connected in tandem, each of said modules comprising a one-dimensional array of processing elements,
- a first input for inputting pixel values from a block of pixels in a current video frame in a predetermined sequence to said modules,
- a second input for inputting pixel values from a search window of a previous video frame in a predetermined sequence to said modules,
- a control circuit for generating a mask signal for each of said modules, and
- a gate circuit for receiving said pixels from said first and second inputs and said mask signals for selectively filtering the transmission of particular pixels inputted at said inputs to particular ones of said modules wherein said current block comprises N.times.N pixels, said search window comprises (2p+1+N).sup.2 pixels, where N=2p+2, and each of said modules comprises 2p+2 processing elements.
A full search block matching circuit comprising:
- a plurality of m modules connected in tandem, each of said modules comprising a one dimensional systolic array, and
- a first pixel input for transmitting to said m modules pixels C from a current block of a current video frame,
- a second pixel input for transmitting to said m modules pixels P from a left-side area of a search window or previous frame having said left-side area and a right-side area,
- a third pixel input for transmitting to said m modules pixels P' from the right-side area of said search window or a previous frame, wherein said pixels C, P and P' are transmitted in a predetermined sequence for enabling said plurality of m modules to determine the errors for m rows of positions of the current block in the search window in parallel.
A full search block matching circuit for a current block of pixels having N pixels in each row of a current video frame in a search window of a previous video frame having {N+(2n+2)p+2n+1} pixels per row, where n.gtoreq.1, and p.gtoreq.1 comprising:
- a plurality of n+1 modules connected in tandem, each of said n+1 modules comprising a one-dimensional systolic array of 2p+2 processing elements for computing an error function,
- a first input for inputting said pixels of said current block to said modules, and
- a second input comprising n+2 distinct inputs for inputting said search window pixels to said modules, each of said distinct inputs inputting pixels from a distinct portion of said search window.
A full search block matching circuit comprising:
- a first module comprising a one-dimensional systolic array of processing elements which is able by itself to compute in parallel the error functions for one row of positions of a current block of pixels of a current video frame in a search window of a first predetermined size of a previous video frame,
- a second module comprising a one-dimensional systolic array of processing elements connected in tandem to said first module, said first and second modules cooperating to compute in parallel one row of positions of said current block in a search window of a second predetermined size larger than said first predetermined size having a left-side area, a middle area and a right-side area, and
- wherein said circuit includes a first input for inputting pixels of said current block to said modules and second, third, and fourth inputs for inputting, to said modules, pixels from the left-side area, the middle area, and the right-side area of said search window of said second size, respectively.
A full search block matching circuit comprising:
- a first module comprising a one-dimensional systolic array which is able by itself to compute in parallel the error functions for one row of positions of a current block of pixels of a current video frame in a search window for a first predetermined size of a previous video frame, and
- a second module comprising a one-dimensional systolic array of processing elements connected in tandem to said first module, said first and second modules cooperating to compute in parallel one row of positions of said current block in a search window of a second predetermined size larger than said first predetermined size,
- wherein said circuit includes a first input for inputting pixels of said current block to said modules and includes second, third, and fourth inputs for inputting to said modules pixels from three distinct areas of said search window of said second size, and wherein each systolic array comprises 2p+2 processing elements, where p.gtoreq.1, said current block has N pixels in each row, where N=2p+2, said search window of a first predetermined size has 2p+1+N pixels in each row, and
- said search window of a second size has 2p+1+N+2(p+1) pixels in each row.
A full search block matching circuit comprising:
- a first module comprising a one-dimensional systolic array which is able by itself to compute in parallel the error functions for one row of positions of a current block of pixels of a current video frame in a search window of a first predetermined size of a previous video frame,
- a second module comprising a one-dimensional systolic array of processing elements connected in tandem to said first module, said first and second modules cooperating to compute in parallel one row of positions of said current block in a search window of a second predetermined size larger than said first predetermined size, and
- one or more additional modules comprising a one-dimensional systolic array connected in tandem to said first and second modules,
- wherein each said additional modules cooperate with said first and second modules to compute in parallel one row of positions of said current block in a search window of a size larger than said second predetermined size,
- wherein said circuit includes a first input for inputting pixels of said current block to said modules and includes second, third, and fourth inputs for inputting to said modules pixels from three distinct areas of said search window of said second size, and wherein each module comprises 2p+2 processing elements, where p.gtoreq.1, and each additional module enables the number of pixels in a row of the search window to be increased by 2(p+1).
A full search block matching circuit comprising:
- a first plurality of modules connected in tandem,
- a second plurality of modules connected in tandem,
- each of said modules in said first and second pluralities comprising a one-dimensional systolic array of processing elements,
- each of said modules including means for receiving pixels of a current block of a current video frame, said pixels of said current block being transmitted serially from one processing element to the next in said first plurality of tandemly connected modules, and being transmitted serially from one processing element to the next in the second plurality of tandemly connected modules, the first processing element of the first module in the second plurality serially receiving said current frame pixels after the last processing element of the first module in the first plurality, and
- each of said modules including means for receiving selected pixels of a search window of a previous video frame at selected times under the control of mask signals.
A full search block matching circuit comprising:
- a first plurality of modules connected in tandem, a second plurality of modules connected in tandem,
- each of said modules in said first and second pluralities comprising a one-dimensional systolic array of processing elements,
- each of said modules including means for receiving pixels of a current block of a current video frame, said pixels of said current block being transmitted serially from one processing element to the next in said first plurality of tandemly connected modules, and being transmitted serially from one processing element to the next in the second plurality of tandemly connected modules, the first processing element of the first module in the second plurality serially receiving said current frame pixels after the last processing element of the first module in the first plurality,
- each of said modules including first, second, and third inputs for receiving pixels from first, second and third different portions of said search window, wherein search window pixels from said first and second inputs are selectively transmitted to the modules in the first plurality under the control of said mask signals, and search window pixels from said second and third inputs are selectively transmitted to said modules of said second plurality under the control of said mask signals, and
- each of said modules determining in parallel the error functions of a plurality of rows of positions of the current block in said search window wherein said current block comprises N pixels in each row, each of said modules comprises 2p+2 processing elements, wherein p.gtoreq.1, and
- said second plurality of modules enabling the number of pixels in a row of the search window to be increased by 2(p+1).
A full search block matching circuit comprising:
- a plurality of m(m>1) modules connected in tandem, each of said modules comprising a one-dimensional array of processing elements,
- a first input for inputting pixel values from a block of pixels in a current video frame in a predetermined sequence to said modules,
- second, third and fourth inputs for inputting pixel values from a respective left-side area, middle area and right-side area of a search window of a previous video frame in a predetermined sequence to said modules,
- a control circuit for generating a stream of mask signals for each of said modules, and
- at least one gate circuit for receiving said pixels from said first and second inputs and said mask signals, and for selectively transmitting particular pixels inputted at said inputs to particular ones of said modules such that each one of said modules being enabled for equal, staggered periods of time.
A full search block matching circuit comprising:
- a first plurality of m(m>1) modules connected in tandem,
- a second plurality of m modules connected in tandem,
- each of said modules in said first and second pluralities comprising a one-dimensional systolic array of processing elements,
- an input for transmitting, to said modules in said first and second pluralities of m modules, pixels from a current block of a current video frame and pixels from a search window of a previous video frame,
- the modules of said first and second pluralities of modules determining in parallel error functions for m rows of positions of the current block in the search window,
- wherein each module comprises 2p+2 processing elements, where p.gtoreq.1, and said second plurality of modules enabling the number of pixels in each row of the search window to increase by 2(p+1).
Dependencies are inferred from the claim text.