Hierarchical randomized quantization of multi-dimensional features
View Patent ↗Systems and methods for quantizing a local descriptor in video fingerprinting applications are provided. In one or more embodiments, local features of a video are extracted and characterized by a set of feature dimensions. The feature dimensions are then quantized to yield a quantized local descriptor for the video. To introduce a degree of pseudorandom variation in the quantization grids, a cascaded random quantization technique is employed to quantize the dimensions, wherein a quantized value for a given dimension is used to quantize a next dimension in sequence.
1. A method, comprising:
calculating a first quantization grid as a function of a first pseudorandom number calculated based on a seed value;
quantizing a first dimension of a descriptor in a fingerprint of a video using the first quantization grid to yield a first quantized value; and
quantizing a second dimension of the descriptor using a second quantization grid calculated as a function of the first quantized value to yield a second quantized value.
2. The method of claim 1 , further comprising: calculating the second quantization grid as a function of a second pseudorandom number calculated based on the seed value and the first quantized value.
3. The method of claim 2 , further comprising: concatenating the first quantized value, the second quantized value, and the seed value to yield a quantized descriptor.
4. The method of claim 2 , further comprising:
calculating a third quantization grid as a function of a third pseudorandom number generated based on the seed value, the first quantized value, and the second quantized value;
quantizing a third dimension of the descriptor using the third quantization grid to yield a third quantized value.
5. The method of claim 1 , wherein the seed value is randomly generated.
6. The method of claim 1 , further comprising: generating multiple quantized values for the first and second dimensions respectively using multiple seeds.
7. A system, comprising:
a memory; and
a processor that executes the following computer-executable components stored within the memory:
a feature extraction component configured to identify N dimensions of a descriptor in a fingerprint of a video, wherein N is an integer greater than 1; and
a quantization component configured to sequentially generate respective N quantized values for the N dimensions, wherein the quantization component quantizes an Mth dimension of the N dimensions using a quantization grid calculated as a function of previously generated quantized values of 1 to M−1 dimensions of the N dimensions, wherein M is an integer from 1 to N.
8. The system of claim 7 , wherein the quantization component is further configured to calculate the quantization grid as a function of a pseudorandom number generated based on a seed value and the previously generated quantized values of the 1 to M−1 dimensions.
9. The system of claim 8 , further comprising: a descriptor generation component configured to concatenate the N quantized values and the seed value to yield a quantized descriptor.
10. The system of claim 8 , wherein the quantization component is further configured to generate multiple quantized values for at least one of the N dimensions by varying the seed value.
11. The system of claim 7 , wherein the descriptor characterizes a video frame of the video.
12. The system of claim 7 , wherein a dimension of the N dimensions includes one of a type, a scale, a position, or a time associated with the descriptor of the video.
13. A non-transitory computer-readable medium having stored thereon computer-executable instructions that, in response to execution, cause a system including a processor to perform operations, comprising:
calculating a first quantization grid as a function of a first pseudorandom number calculated based on a seed value;
quantizing a first dimension of a descriptor in a fingerprint of a video using a first quantization grid to yield a first quantized value; and
quantizing a second dimension of the descriptor using a second quantization grid calculated as a function of the first quantized value to yield a second quantized value.
14. The non-transitory computer-readable medium of claim 13 , the operations further comprising: calculating the second quantization grid as a function of a second pseudorandom number calculated based on the seed value and the first quantized value.
15. The non-transitory computer-readable medium of claim 14 , the operations further comprising:
calculating a third quantization grid as a function of a third pseudorandom number generated based on the seed value, the first quantized value, and the second quantized value;
quantizing a third dimension of the descriptor using the third quantization grid to yield a third quantized value.
16. The non-transitory computer-readable medium of claim 13 , the operations further comprising: concatenating the first quantized value, the second quantized value, and the seed value to yield a quantized descriptor.
17. The non-transitory computer-readable medium of claim 13 , wherein the descriptor is at least one interest point.
18. The non-transitory computer-readable medium of claim 13 , the operations further comprising: generating multiple quantized values for the first and second dimensions respectively by using multiple seeds.