Image down-scaling with pixel sets selected via blue noise sampling
In some embodiments, a computing device uses a blue noise sampling operation to identify source pixels from an input image defining respective pixel sets. Each pixel set is associated with a respective weight matrix for a down-scaling operation. The blue noise sampling operation causes an overlap region between first and second pixel sets. The computing device assigns an overlap pixel in the overlap region to the first weight matrix based on the overlap pixel being closer to the first source pixel. The computing device modifies the second weight matrix to exclude the overlap pixel from a portion of the down-scaling operation involving the second weight matrix. The computing device performs the down-scaling operation on the input image by combining the first pixel set into a first target pixel with the first weight matrix and combining the second pixel set into a second target with the modified second weight matrix.
1. A method that includes one or more processing devices performing operations comprising:
identifying, by applying a blue noise sampling operation to an input image, a first source pixel associated with a first set of pixels in the input image and a second source pixel associated with a second set of pixels in the input image, wherein the first set of pixels and the second set of pixels have an overlap pixel, wherein the first set of pixels is associated with a first weight matrix and the second set of pixels is associated with a second weight matrix;
assigning the overlap pixel to the first set of pixels;
modifying the second weight matrix to exclude the overlap pixel from a down-scaling operation for with the second set of pixels; and
transforming the input image into a down-scaled image by at least performing the down-scaling operation, the down-scaling operation comprising combining the first set of pixels using the first weight matrix and combining the second set of pixels using the second weight matrix as modified.
2. The method of claim 1 , wherein applying the blue noise sampling operation further comprises:
receiving a parameter indicating a degree of randomness for the blue noise sampling operation; and
modifying the blue noise sampling operation based on the parameter.
3. The method of claim 2 , wherein the blue noise sampling operation maintains a distance between the first source pixel and the second source pixel that exceeds a threshold distance, wherein the parameter comprises the threshold distance.
4. The method of claim 1 , wherein performing the down-scaling operation further comprises performing (i) a first convolution operation on the first weight matrix and the first set of pixels and (ii) a second convolution operation on the second weight matrix, as modified, and the second set of pixels.
5. The method of claim 1 , wherein assigning the overlap pixel to the first set of pixels is performed in response to determining that the overlap pixel is closer to the first source pixel than the second source pixel.
6. The method of claim 1 , wherein modifying the second weight matrix further comprises:
determining a bilateral energy based on a position of the first source pixel, a position of the second source pixel, the first weight matrix, and the second weight matrix; and
modifying the second weight matrix to minimize the bilateral energy.
7. The method of claim 6 , wherein minimizing the bilateral energy comprises minimizing a difference between the first weight matrix and the second weight matrix for the overlap pixel.
8. A system comprising:
a processing device; and
a non-transitory computer-readable medium communicatively coupled to the processing device, wherein the processing device is configured to execute program code stored in the non-transitory computer-readable medium and thereby perform operations comprising:
identifying, by applying a blue noise sampling operation to an input image, a first source pixel associated with a first set of pixels in the input image and a second source pixel associated with a second set of pixels in the input image, wherein the first set of pixels and the second set of pixels have an overlap pixel, wherein the first set of pixels is associated with a first weight matrix and the second set of pixels is associated with a second weight matrix;
assigning the overlap pixel to the first set of pixels;
modifying the second weight matrix to exclude the overlap pixel from a down-scaling operation for with the second set of pixels; and
transforming the input image into a down-scaled image by at least performing the down-scaling operation, the down-scaling operation comprising combining the first set of pixels using the first weight matrix and combining the second set of pixels using the second weight matrix as modified.
9. The system of claim 8 , wherein applying the blue noise sampling operation further comprises:
receiving a parameter indicating a degree of randomness for the blue noise sampling operation; and
modifying the blue noise sampling operation based on the parameter.
10. The system of claim 9 , wherein the blue noise sampling operation maintains a distance between the first source pixel and the second source pixel that exceeds a threshold distance, wherein the parameter comprises the threshold distance.
11. The system of claim 8 , wherein performing the down-scaling operation further comprises performing (i) a first convolution operation on the first weight matrix and the first set of pixels and (ii) a second convolution operation on the second weight matrix, as modified, and the second set of pixels.
12. The system of claim 8 , wherein assigning the overlap pixel to the first set of pixels is performed in response to determining that the overlap pixel is closer to the first source pixel than the second source pixel.
13. The system of claim 8 , wherein modifying the second weight matrix further comprises:
determining a bilateral energy based on a position of the first source pixel, a position of the second source pixel, the first weight matrix, and the second weight matrix; and
modifying the second weight matrix to minimize the bilateral energy.
14. The system of claim 13 , wherein minimizing the bilateral energy comprises minimizing a difference between the first weight matrix and the second weight matrix for the overlap pixel.
15. A non-transitory computer-readable medium having program code that is stored thereon, the program code executable by one or more processing devices for performing operations comprising:
identifying, by applying a blue noise sampling operation to an input image, a first source pixel associated with a first set of pixels in the input image and a second source pixel associated with a second set of pixels in the input image, wherein the first set of pixels and the second set of pixels have an overlap pixel, wherein the first set of pixels is associated with a first weight matrix and the second set of pixels is associated with a second weight matrix;
assigning the overlap pixel to the first set of pixels;
modifying the second weight matrix to exclude the overlap pixel from a down-scaling operation for with the second set of pixels; and
transforming the input image into a down-scaled image by at least performing the down-scaling operation, the down-scaling operation comprising combining the first set of pixels using the first weight matrix and combining the second set of pixels using the second weight matrix as modified.
16. The non-transitory computer-readable medium of claim 15 , wherein applying the blue noise sampling operation further comprises:
receiving a parameter indicating a degree of randomness for the blue noise sampling operation; and
modifying the blue noise sampling operation based on the parameter.
17. The non-transitory computer-readable medium of claim 16 , wherein the blue noise sampling operation maintains a distance between the first source pixel and the second source pixel that exceeds a threshold distance, wherein the parameter comprises the threshold distance.
18. The non-transitory computer-readable medium of claim 15 , wherein performing the down-scaling operation further comprises performing (i) a first convolution operation on the first weight matrix and the first set of pixels and (ii) a second convolution operation on the second weight matrix, as modified, and the second set of pixels.
19. The non-transitory computer-readable medium of claim 15 , wherein assigning the overlap pixel to the first set of pixels is performed in response to determining that the overlap pixel is closer to the first source pixel than the second source pixel.
20. The non-transitory computer-readable medium of claim 15 , wherein modifying the second weight matrix further comprises:
determining a bilateral energy based on a position of the first source pixel, a position of the second source pixel, the first weight matrix, and the second weight matrix; and
modifying the second weight matrix to minimize the bilateral energy, wherein minimizing the bilateral energy comprises minimizing a difference between the first weight matrix and the second weight matrix for the overlap pixel.