Method for stochastic computing image processing using correlation controlled contingency tables

1 . A method for processing an input image in a stochastic computing system utilizing at least one multi-level cascaded stochastic computing circuit comprising:

a. establishing one or more contingency tables comprising a 2×2 matrix comprising four primitives, a, b, c, and d;

b. storing the contingency table in a memory in the stochastic computing circuit; and

c. processing the input image, comprising the following:

i. combining the input image with an additional image through an image composition process, wherein the image composition process comprises:

aa. providing the input image, comprising a background image (B);

bb. providing an input foreground image (F);

cc. providing a foreground image opacity of the input foreground image (a);

dd. providing a two-input multiplexer;

ee. identifying a probability value of the background image P B ;

ff. identifying a probability value of the foreground image P F ;

gg. identifying a probability value of the foreground image opacity P a ; and

hh. generating a composite image by performing the following operations: P B ×(1−P a )+(P F ×P a ),

ii. identifying a degree of similarity between the input image and an additional image through a template matching process; and

iii. comparing the input image with a comparison image through a pattern comparison process.

2 . The method of claim 1 , wherein the template matching process comprises:

a. determining a size of the input image, comprising a total amount of pixels, r×c;

b. providing a template, K, wherein said template comprises a kernel that is a subset of the input image;

c. comparing each pixel of the input image with the template;

d. identifying each pixel where the input image pixel and template intersect;

e. generating a temporary matching image; and

f. normalizing the temporary matching image.

3 . The method of claim 2 , wherein generating the temporary matching image comprises by applying Σ f(I, K), wherein/represents the input image; and

wherein f(I, K) comprises any one of the following:

a. |I-K|;

b. (I-K) 2 ; and

c. I×K.

4 . The method of claim 2 , further comprising:

a. generating a contingency table for the input image;

b. generating a contingency table for the template;

c. generating the temporary matching image by performing an AND operation on the input image and template, wherein the input image and the template each comprise zero correlated bit streams;

d. decoding the temporary matching image by identifying a population count of 1s, comprising a result ΣI×K;

e. subtracting K 2 from the decoding result; and

f. identifying an exact template match wherein a result of the foregoing steps generates a zero result.

5 . The method of claim 2 , further comprising:

a. generating a contingency table for the input image;

b. generating a contingency table for the template;

c. generating the temporary matching image by performing a stochastic computing multiplication operation on the input image and template, wherein the input image and the template each comprise zero correlated bit streams;

d. decoding the temporary matching image by identifying a population count of 1s; and

e. identifying a template match wherein a result of the foregoing steps generates a zero result.

6 . The method of claim 2 , further comprising:

a. generating a contingency table for the input image;

b. generating a contingency table for the template;

c. generating the temporary matching image by performing an XOR operation on the input image and template, wherein the input image and the template each comprise maximally correlated bit streams;

d. decoding the temporary matching image by identifying a population count of 1s; and

e. identifying a template match wherein a result of the foregoing steps generates a zero result.

7 . The method of claim 1 , wherein the pattern comparison process comprises:

a. calculating a maximum a value by applying min (M, P), wherein M represents the input image and P represents the comparison image;

b. setting the maximum a value as an expected a value;

c. calculating b, comprising subtracting a from the input image;

d. calculating c, comprising subtracting a from the comparison image;

e. applying an XOR operation to b and c; and

f. determining any differences between the input image and comparison image;

wherein a zero value is obtained when an exact pattern match is present.

8 . The method of claim 1 , wherein the pattern comparison process comprises:

a. generating the contingency table comprising:

i. providing a plurality of inputs to the contingency table comprising a bit stream length (N), at least two input scalars X1 and X2, and a target correlation;

ii. establishing a range of correlation points for a, wherein said range comprises a minimum a value, a maximum a value, and a zero a value;

iii. calculating the maximum a value by applying min (M, P), wherein M represents the input image and P represents the comparison image;

b. setting the maximum a value as an expected a value;

c. calculating b, comprising subtracting a from the input image;

d. calculating c, comprising subtracting a from the comparison image;

e. applying an XOR operation to b and c; and

f. determining any differences between the input image and comparison image;

wherein a zero value is obtained when an exact pattern match is present.