Method and system for encoding image data in hyperdimensional computing systems
A method and architecture for encoding image data solely based on pixel intensities, eliminating the need for encoding pixel positions. This is achieved by utilizing low-discrepancy (LD) sequences, such as Sobol or Van Der Corput (VDC) sequences, for deterministic encoding of the image intensity values. The position hypervectors, which were previously required for representing pixel positions, are no longer needed. The corresponding index of the LD sequence is used to encode the intensity values.
1 . A method for encoding an image data comprising the following steps:
(a) providing hyperdimensional computing system comprising:
a low-discrepancy sequence generator;
a comparator;
a pop-counter;
a subtractor; and
a memory;
(b) identifying the image data to be encoded, wherein the image data comprises a plurality of pixels;
(c) generating a low-discrepancy sequence by the low-discrepancy sequence generator, wherein the low-discrepancy sequence comprises a plurality of numbers;
(d) for each pixel of the image data, identifying a normalized pixel intensity;
(e) comparing each number of the low-discrepancy sequence to a corresponding normalized pixel intensity value for each pixel, wherein:
(i) if the normalized pixel intensity is greater than a corresponding number of the low-discrepancy sequence, a hypervector position of said pixel is assigned a +1; and
(ii) if the normalized pixel intensity is less than a corresponding number of the low-discrepancy sequence, a hypervector position of said pixel is assigned a −1;
(iii) repeating the above-described comparison for all pixels in the image data;
(f) accumulating the hypervector positions from the pixels; and
(g) saving a final encoded image data to the memory.
2 . The method of claim 1 , wherein the low-discrepancy sequences comprise one or more Sobol sequences.
3 . The method of claim 1 , wherein the low-discrepancy sequences comprise one or more Halton sequences.
4 . The method of claim 1 , wherein the low-discrepancy sequences comprise one or more Van der Corput sequences.
5 . The method of claim 1 , wherein the low-discrepancy sequences comprise a combination of two or more of the following:
Sobol sequences;
Halton sequences; and
Van der Corput sequences.
6 . The method of claim 1 , wherein the steps for encoding the image data are conducted using an ARM-based embedded computing platform.
7 . The method of claim 1 , wherein the low-discrepancy sequence generator comprises a log 2 D-bit counter.
8 . The method of claim 1 , wherein the comparison is performed using a cosine similarity process.
9 . A system comprising at least one computing device comprising:
a low-discrepancy sequence generator;
a comparator;
a pop-counter;
a subtractor; and
a memory;
wherein the computing device is configured to:
receive image data comprising a plurality of pixels, wherein each pixel comprises a normalized pixel intensity value;
compare one or more low-discrepancy sequence values generated by the low-discrepancy sequence generator to a normalized pixel intensity value for each pixel;
accumulate a hypervector position from each pixel; and
saving a final encoded image data to the memory.
10 . The system of claim 9 , wherein the low-discrepancy sequence values comprise one or more Sobol sequences values.
11 . The system of claim 9 , wherein the low-discrepancy sequence values comprise one or more Halton sequences values.
12 . The system of claim 9 , wherein the low-discrepancy sequence values comprise one or more Van der Corput sequence values.
13 . The system of claim 9 , wherein the low-discrepancy sequence values comprise a combination of two or more of the following low-discrepancy sequences:
Sobol sequences;
Halton sequences; and
Van der Corput sequences.
14 . The system of claim 9 , wherein the computing device comprises an ARM-based embedded computing platform.
15 . The system of claim 9 , wherein the low-discrepancy sequence generator comprises a log 2 D-bit counter.
16 . The system of claim 9 , wherein the comparison is performed using a cosine similarity process.