Image processing apparatus

1. An image processing apparatus comprising a normalisation module operatively connected across a bus to a memory storing an image in which a region of interest (ROI) has been identified within the image, the ROI being bound by a rectangle having a non-orthogonal orientation within the image, the normalisation module being arranged to:

divide said ROI into one or more slices, each slice comprising a plurality of adjacent rectangular tiles;

for each slice, successively read ROI information for each tile from said memory including: reading a portion of said image extending across at least a width of said slice line-by-line along an extent of a slice;

for each tile, downsample said ROI information to a buffer to within a scale S D <2 of a required scale for a normalised version of said ROI; and fractionally downsample and rotate downsampled information for a tile within said buffer to produce a respective normalised portion of the ROI at the required scale for the normalised ROI; and

accumulate downsampled and rotated information for each tile within a normalised ROI buffer for subsequent processing by said image processing apparatus.

2. An image processing apparatus according to claim 1 wherein said ROI is divided into S slices and each slice is divided into T=S tiles.

3. An image processing apparatus according to claim 1 wherein said ROI is oriented at one of a fixed number of possible non-orthogonal orientations and the direction of reading successive slices and the order of reading successive tiles is a function of said orientation.

4. An image processing apparatus according to claim 3 wherein said fixed number is 8 and said orientations are evenly spaced from one another to detect objects at orientations within respective half-quadrant ranges of orientation.

5. An image processing apparatus according to claim 4 wherein successive tiles from ROI in adjacent half-quadrants are read in opposite directions.

6. An image processing apparatus according to claim 4 wherein successive slices from ROI in opposite hemispheres are read in opposite directions.

7. An image processing apparatus according to claim 3 wherein information for a slice is read from said memory beginning at a memory address corresponding to a leftmost pixel of a slice.

8. An image processing apparatus according to claim 1 wherein said portion of said image read from memory is constant for each line of a slice.

9. An image processing apparatus according to claim 8 wherein a start memory address for reading each line of a slice is advanced according to the non-orthogonal orientation of said ROI.

10. An image processing apparatus according to claim 8 wherein said advance is non-linear for a first tile of a slice.

11. An image processing apparatus according to claim 8 wherein said advance is linear for second and subsequent tiles of a slice.

12. An image processing apparatus comprising a normalisation module operatively connected across a bus to a memory storing an image in which a region of interest (ROI) has been identified within the image, the ROI being bound by a rectangle having a non-orthogonal orientation within the image, the normalisation module being arranged to:

read ROI information from said memory including: reading a portion of said image extending across at least a width of said ROI line-by-line along an extent of the ROI;

for successive sets of lines of said ROI, downsample said ROI information to a buffer to within a scale S D <2 of a required scale for a normalised version of said ROI;

for each set of lines, fractionally downsample and rotate downsampled information for the set of lines within said buffer to produce a respective normalised line of the ROI at the required scale for the normalised ROI; and

accumulate downsampled and rotated lines within a normalised ROI buffer for subsequent processing by said image processing apparatus.

13. An image processing apparatus according to claim 12 wherein information for a line is read from said memory beginning at a start memory address corresponding to a leftmost pixel of a line and extending to an end memory address corresponding to a rightmost pixel of a line.

14. An image processing apparatus according to claim 12 wherein said apparatus is arranged to calculate a shift in said start memory address and said end memory address according to the non-orthogonal orientation of said ROI.

15. An image processing apparatus according to claim 14 wherein said apparatus is arranged to recalculate said shifts at any line incorporating a corner of said ROI.

16. An image processing apparatus according to claim 12 wherein a number of lines in said buffer into which said downsampled information is written is a function of the orientation of said ROI.

17. An image processing apparatus according to claim 1 wherein said apparatus is arranged to selectively flip information for said ROI according to said non-orthogonal orientation so that objects within said normalised ROI buffer have a common orientation.

18. An image processing apparatus according to claim 1 wherein said ROI of interest comprises a face detected within said image.

19. An image processing apparatus according to claim 1 wherein said apparatus is further arranged to perform one or both of histogram equalisation or contrast enhancement of the information within said normalised ROI buffer.

20. An image processing apparatus according to claim 1 wherein said apparatus is further arranged to determine a mean intensity value of the information within said normalised ROI buffer.

21. An image processing apparatus according to claim 1 wherein said apparatus is further arranged to generate one or more of the following maps based on the information within said normalised ROI buffer:

an Integral Image (II)

an II 2 map

a Census map

a LBP (linear binary pattern)

a HOG (Histogram of Gradients) and

a Convolution of maps generated from a plurality of normalised ROI generated from respective images comprising similar content.

22. An image processing apparatus according to claim 1 wherein the normalisation module is responsive to the ROI being bound by a rectangle having an orthogonal orientation within the image to:

divide said ROI into one or more rectangular tiles; and

for each tile, successively read ROI information for each tile from said memory including:

reading a portion of said image extending across at least a width of said tile; and

downsampling information for a tile to produce a respective normalised portion of the ROI at the required scale for the normalised ROI; and

accumulating downsampled information for each tile within a normalised ROI buffer for subsequent processing by said image processing apparatus.