# US 2002/0154123 A1 · Image scaling

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2002/0154123 A1
- Document ID: US20020154123A1
- Status: Published application
- Kind: A1
- Application number: 10/050,880
- Filed: 2002-01-18
- Published: 2002-10-24
- Earliest claimed: 2001-01-19
- Assignees: Clearspeed Technology Limited
- Inventors: Michal Harasimiuk
- Web page: https://patentgrep.com/patent/US20020154123A1

## CPC

- G06T 3/4023: based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A method and a system for scaling a digital source image consisting of a grid of X by Y pixels into a target image of a different resolution comprises the steps of/tools for: mapping the source pixels onto the target pixels; scaling the source image in the X or Y direction to produce intermediate pixels that are scaled in one direction by determining contributions to each intermediate pixel using a digital filter and accumulating the contributions for each intermediate pixel, wherein each source pixel contributes to one or more intermediate pixels and each intermediate pixel receives contributions from one or more source pixels; and subsequently scaling the intermediate pixels in the other direction by determining the contributions to each target pixel using the filter and accumulating the contributions for each target pixel; wherein each intermediate pixel contributes to one or more target pixels and each target pixel receives contributions from one or more intermediate pixels. The method is preferably carried out on a PE array in an SIMD (Simple Instruction Multiple Data) device.

## Claims

### Claim 1

A parallel processing method of scaling a digital source image consisting of a matrix of X by Y pixels into a target image, comprising the steps of:
mapping the source pixels onto the target pixels;
scaling the source image in the X or Y direction in parallel to produce intermediate pixels that are scaled in one direction by determining contributions to each intermediate pixel using a digital filter and accumulating the contributions for each intermediate pixel, wherein each source pixel contributes to one or more intermediate pixels and each intermediate pixel receives contributions from one or more source pixels; and subsequently
scaling the intermediate pixels in the other direction in parallel to produce target pixels by determining the contributions to each target pixel using the filter and accumulating the contributions for each target pixel;
wherein each intermediate pixel contributes to one or more target pixels and each target pixel receives contributions from one or more intermediate pixels.

### Claim 2 (depends on 1)

A method according to claim 1, including the step of mapping a cluster of adjacent higher-resolution pixels onto each of the lower resolution pixels, in which a higher resolution pixel belongs to the cluster of a lower resolution pixel if its centre falls within that lower resolution pixel.

### Claim 3 (depends on 1)

A method according to claim 1, including the step of calculating the distance between the centre of each higher resolution pixel and the centre of each lower resolution pixel to which it contributes or from which it receives contributions in the X and Y directions.

### Claim 4 (depends on 1)

A method according to claim 1, which integral filter function values for the filter function are determined for the X direction in an X preparation step, preferably prior to reading the pixel value data.

### Claim 5 (depends on 2)

A method according to claim 2, including the step of calculating the distance from the cluster boundary as the source pixels are read in and defining process changes or increments to occur when the cluster border is crossed.

### Claim 6 (depends on 1)

A method according to claim 1, in which the filter function for each lower resolution pixel is evaluated at any suitable number of points known as the filter footprint number, each point corresponding to a source pixel in downscaling or to a target pixel in upscaling; the filter footprint number being determined by the footprint of the filter function in source space or target space respectively.

### Claim 7 (depends on 1)

A method according to claim 1, in which the filter function is an integration function which is determined analytically, without approximation of the integration.

### Claim 8 (depends on 1)

A method according to claim 1, in which in downscaling, (and in each direction) each source pixel contributes to a number of target pixels determined by the filter width and in upscaling (and in each direction) each target pixel receives contributions from a number of source pixels determined by the filter width.

### Claim 9 (depends on 1)

A method according to claim 1, in which a processing element (PE) array is used, in which interconnections allow data to be moved (swazzled) along the PE array, each PE including a number of addressable memory cells, in which values corresponding to one or more pixels may be stored, and between which the values may be shifted.

### Claim 10 (depends on 9)

a method according to claim 9 in which one PE is provided per pixel of the higher resolution image and the memory cell array is preferably of the same length as the filter width.

### Claim 11 (depends on 9)

A method according to claim 9 in which one PE is used per pixel of the lower resolution image and each PE processes two or more neighbouring higher resolution pixels which are read into the PE sequentially.

### Claim 12 (depends on 9)

A method according to claim 9, in which a PE-pixel mapping is automatically selected according to process conditions.

### Claim 13 (depends on 1)

A method according to claim 1, using SIMD processing on each line of source pixels.

### Claim 14 (depends on 9)

A method according to claim 9, in which values are shifted in the memory cell array during Y scaling such that they are shifted up the array one position when the process crosses a cluster boundary.

### Claim 15 (depends on 9)

A method according to claim 9, in which data is swazzled between PEs for X scaling; the data being swazzled a certain distance across a predetermined number of cluster boundaries until it reaches the correct PE for accumulation to make up target pixels.

### Claim 16 (depends on 1)

A method according to claim 1, including the steps of reading the source image before scaling and analysing the source and target dimensions and writing the target pixels to external memory after scaling.

### Claim 17 (depends on 9)

A method according to claim 9, in which the Y scaling step for downscaling includes the steps of multiplying the source pixel by the appropriate filter function values and reading the resultant contributions for each source pixel into the cells in the memory cell array, adding them to any contributions from one or more lines above the present scanline which are already in the cell; and incrementing to shift the feed into the array by one position and move the top cell value out when a cluster border is crossed.

### Claim 18 (depends on 9)

A method according to claim 9, in which the Y scaling step for upscaling includes the steps of reading each pixel value into a cell, calculating the contributions for each pixel in the array using the filter function; summing the contributions (in a number corresponding to the filter width) for each target pixel; and shifting the source pixel values up one position, to read in a new source pixel and discard an old source pixel when the target boundary is crossed.

### Claim 19 (depends on 9)

A method according to claim 9, in which X scaling involves accumulating contributions in one accumulating PE per cluster, the contributions for each source pixels being swazzled across the PE array to their correct accumulating PE.

### Claim 20 (depends on 9)

A method according to claim 9, in which the method for the X scaling step for downscaling calculates the contributions for each source pixel; addresses them according to the distance they must travel and stores them in a memory cell array, then shifts each contribution in turn along the PE array to the correct accumulating PE.

### Claim 21 (depends on 9)

A method according to claim 9, in which the method for the X scaling stop for upscaling involves loading the PE array with a repeating sequence of the source pixels in the corresponding cluster of the target pixels.

### Claim 22 (depends on 21)

A method according to claim 21 in which the sequence is indexed to increase by one and restart each time a cluster boundary is crossed, the sequence restarting with the first pixel in the sequence and having a length equal to the filter width.

### Claim 23 (depends on 21)

A method according to claim 21 in which the value loaded is held in one cell of the array of memory cells and the preferred swazzle movement moves the values one step at a time to fill a different cell with its value.

### Claim 24 (depends on 1)

A computer program, algorithm or method of controlling a computer for use in SIMD processing to scale a digital source image consisting of a matrix of X by Y pixels into a target image using a method as defined in claim 1.

### Claim 25 (depends on 24)

A computer program product comprising the program according to claim 24 loaded on a carrier, such as a CD or ROM memory or transported on a carrier wave.

### Claim 26 (depends on 1)

A parallel data processing apparatus for scaling a digital source image consisting of a matrix of X by Y pixels into a target image, such as a computer or set-top box, image processing apparatus, graphics or multi-media system comprising a PE array as hereinbefore described and calculating tools for carrying out the method as defined in claim 1.

### Claim 27

A parallel processing system for scaling a digital source image consisting of a matrix of X by Y pixels into a target image, comprising:
a mapping tool for mapping the source pixels onto the target pixels;
a first parallel processing scaling tool for scaling the source image in the X or Y direction to produce intermediate pixels that are scaled in one direction by determining contributions to each intermediate pixel using a digital filter and accumulating the contributions for each intermediate pixel, wherein each source pixel contributes to one or more intermediate pixels and each intermediate pixel receives contributions from one or more source pixels; and
a second parallel processing scaling tool for scaling the intermediate pixels in the other direction by determining the contributions to each target pixel using the filter and accumulating the contributions for each target pixel;
wherein each intermediate pixel contributes to one or more target pixels and each target pixel receives contributions from one or more intermediate pixels.
