System and method for efficient image feature extraction
View Patent ↗A system for efficient image feature extraction comprises a buffer for storing a slice of at least n lines of gradient direction pixel values of a directional gradient image. The buffer has an input for receiving the first plurality n of lines and an output for providing a second plurality m of columns of gradient direction pixel values of the slice to an input of a score network, which comprises comparators for comparing the gradient direction pixel values of the second plurality of columns with corresponding reference values of a reference directional gradient pattern of a shape and adders for providing partial scores depending on output values of the comparators to score network outputs which are coupled to corresponding inputs of an accumulation network having an output for providing a final score depending on the partial scores.
1. A system for efficient image feature extraction, comprising
a buffer for storing a slice of at least a first plurality n of lines of gradient direction pixel values of a directional gradient image, said buffer having
a buffer input for receiving said first plurality n of lines and
a buffer output for providing a second plurality m of columns of gradient direction pixel values of said slice to a score network input of a score network, said score network comprising:
a plurality of comparators for comparing said gradient direction pixel values of said second plurality m of columns with corresponding reference gradient direction pixel values of a reference directional gradient pattern of a shape; and
a plurality of adders for providing partial score values depending on output values of said comparators to one or more score network outputs, said one or more score network outputs coupled to one or more corresponding accumulation network inputs of
an accumulation network having
an accumulation network output for providing a final score value depending on said partial score values.
2. The system as claimed in claim 1 , wherein said buffer is a ring buffer.
3. The system as claimed in claim 1 , wherein said buffer allows for storing at least one additional line of gradient direction pixel values.
4. The system as claimed in claim 1 , comprising
a gradient direction calculation module having
a module input for receiving input pixels of an input image and
a module output for providing said gradient direction pixel values of said directional gradient image calculated from said input image to said buffer input.
5. The system as claimed in claim 1 , wherein said comparators and adders are coupled to form a plurality of column comparator networks, each column comparator network providing one of said partial scores to said accumulation network.
6. The system as claimed in claim 5 , wherein said second plurality m of columns of said slice is applied to said score network consecutively and each column of said second plurality m of columns is applied in parallel to each of said column comparator networks.
7. The system as claimed in claim 6 , wherein said accumulation network comprises a plurality of accumulate-and-shift registers coupled to form an accumulator chain.
8. The system as claimed in claim 7 , wherein;
a second column b of the second plurality of m columns is applied to said score network after a first column a of the second plurality of m columns is applied to said score network;
a partial score provided by a column comparator network of the plurality of column comparator networks in the application of column a to said score network is a result C a of a comparison of gradient direction pixel values of column a with corresponding reference gradient direction pixel values of the reference directional gradient pattern;
each of said accumulate-and-shift registers is coupled to a different column comparator network for receiving one of said partial score values;
the accumulate-and-shift registers are for shifting the result C a to another column comparator network of the plurality of column comparator networks; and
a partial score provided by the other column comparator network of the plurality of column comparator networks in the cation of column b to said score network is the sum C a +C b , wherein C b is a result of a comparison of gradient direction pixel values of column b with corresponding reference gradient direction pixel values of the reference directional gradient pattern.
9. The system as claimed in claim 1 , wherein said shape is a circle having a diameter of n or less pixel values.
10. The system of claim 9 , wherein the system does not include a storage area comprising storage for a 2-dimensional set of values of circle-center probability scores.
11. The system as claimed in claim 1 , wherein said gradient direction pixel values and said reference gradient direction pixel values are quantized depending on the same quantization characteristic.
12. The system as claimed in claim 1 , further comprising:
one or more additional score networks, each additional score network for providing score values corresponding to a different shape, coupled to a corresponding additional accumulation network.
13. An image feature extraction method, comprising:
storing a slice of at least a first plurality n of lines of gradient direction pixel values of a directional gradient image in a buffer of a computing device;
providing by the computing device a second plurality m of columns of gradient direction pixel values of said slice to a score network input of a score network;
comparing said gradient direction pixel values of said second plurality m of columns with corresponding reference gradient direction pixel values of a reference directional gradient pattern of a shape;
providing partial score values depending on result values of said comparisons; and
providing a final score value depending on said partial score values.
14. The image feature extraction method as claimed in claim 13 , wherein:
the step of providing partial score values comprises providing each of said partial score values as a result of a comparison of a column of said reference gradient direction pattern and a column of said second plurality m of columns; and
the step of providing the second plurality m of columns of gradient direction pixel values of said slice to a score network input of a score network comprises:
providing each of said second plurality m of columns of said slice to said score network consecutively, and
applying each column of said second plurality m of columns in parallel to each of a plurality of column comparator networks, each application of a column to a column comparator network providing a partial score value of the partial score values, wherein each application represents a different center position of the shape.
15. A computer program product, comprising a computer-readable medium containing code portions for executing steps of a method as claimed in claim 13 .
16. A system-on-a-chip, comprising
a buffer for storing a slice of at least a first plurality n of lines of gradient direction pixel values of a directional gradient image, said buffer having
a buffer input for receiving said first plurality n of lines and
a buffer output for providing a second plurality m of columns of gradient direction pixel values of said slice to a score network input of a score network, said score network comprising:
a plurality of comparators for comparing said gradient direction pixel values of said second plurality m of columns with corresponding reference gradient direction pixel values of a reference directional gradient pattern of a shape; and
a plurality of adders for providing partial score values depending on output values of said comparators to one or more score network outputs, said one or more score network outputs coupled to one or more corresponding accumulation network inputs of
an accumulation network having
an accumulation network output for providing a final score value depending on said partial score values.
17. An advanced driver assistance system (ADAS), comprising
a system-on-a-chip as claimed in claim 16 .
18. A vehicle, comprising
a system-on-a-chip as claimed in claim 16 .
19. The system of claim 1 wherein:
each of the second plurality m of columns of the buffer output includes n elements;
the reference gradient direction pixel values comprise an array with m columns, each column including n elements;
a gradient direction pixel value B i,j of the second plurality m of columns of the buffer output corresponds to a gradient direction pixel value REFi,j of the reference gradient direction pixel values, where B i,j is a gradient direction pixel value of an ith element of a jth column of the m columns of the output buffer and REF i,j is a gradient direction pixel value of an ith element of a jth column of the m columns of the reference gradient direction pixel values; and
the final score value is determined by the formula:
Final
Score
Value
=
∑
i
=
i
,
n
j
=
1
,
m
REF
i
,
j
X
B
i
,
j
where the operator X denotes a comparison between gradient direction pixel values.
20. The system of claim 19 , wherein the comparison X between gradient direction pixel values PV 1 and PV 2 is defined as follows:
PV1×PV2=1, if PV1 and PV2 are both valid and are both of the same gradient direction; and
0, otherwise.