Image classification by convolutional neural network using radial summation
View Patent ↗A method, a computer program product, and a computer system classify an image with a convolutional neural network. The method receiving an image. The method includes performing a radial summation on the image to generate a radially summed image. The method includes inputting the radially summed image into the CNN to perform an image classification.
1. A computer-implemented method for classifying an image with a convolutional neural network (CNN) into an image category, the computer-implemented method comprising:
receiving an image with two spatial dimensions for classification by a convolutional neural network (CNN), the image having an arbitrary rotation in an image plane;
performing a radial summation preprocessing on the image to generate a radially summed image prior to utilizing the CNN for classification, the radial summation converting the image with two spatial dimensions into a tensor with three spatial dimensions without relying upon the CNN to perform the radial summation;
inputting the radially summed image into the CNN to perform an image classification, the image classification classifying the image into an image category; and
classifying the radially summed images into the image category with the CNN.
2. The computer-implemented method of claim 1 , wherein the three spatial dimensions further includes an azimuthal angle.
3. The computer-implemented method of claim 2 , wherein the radial summation comprises:
performing a weighted sum of radial segments starting from a given pixel of the image for a set number of a plurality of the azimuthal angle.
4. The computer-implemented method of claim 1 , wherein the radial summation is performed independently for each image channel.
5. The computer-implemented method of claim 1 , wherein the CNN performs three-dimensional convolutions on the radially summed image.
6. The computer-implemented method of claim 1 , wherein the radial summation comprises:
converting three spatial dimensions of the image to a tensor with one of five spatial dimensions or four spatial dimensions if only one plane of rotation is considered.
7. A non-transitory computer-readable storage media executed by a computer to perform program instructions stored on the non-transitory computer-readable storage media for classifying an image with a convolutional neural network (CNN) into an image category, the non-transitory computer-readable storage media storing the program instructions comprising:
receiving an image with two spatial dimensions for classification by a convolutional neural network (CNN), the image having an arbitrary rotation in an image plane;
performing a radial summation preprocessing on the image to generate a radially summed image prior to utilizing the CNN for classification, the radial summation converting the image with two spatial dimensions into a tensor with three spatial dimensions without relying upon a specialized layer of the CNN to perform the radial summation; and
inputting the radially summed image into the CNN to perform an image classification, the image classification classifying the image into an image category; and
classifying the radially summed images into the image category with the CNN.
8. The non-transitory computer-readable storage media of claim 7 , wherein the three spatial dimensions further includes an azimuthal angle.
9. The non-transitory computer-readable storage media of claim 8 , wherein the radial summation comprises:
performing a weighted sum of radial segments starting from a given pixel of the image for a set number of a plurality of the azimuthal angle.
10. The non-transitory computer-readable storage media of claim 7 , wherein the radial summation is performed independently for each image channel.
11. The non-transitory computer-readable storage media of claim 7 , wherein the CNN performs three-dimensional convolutions on the radially summed image.
12. The non-transitory computer-readable storage media of claim 7 , wherein the radial summation comprises:
converting three spatial dimensions of the image to a tensor with one of five spatial dimensions or four spatial dimensions if only one plane of rotation is considered.
13. A computer system for classifying an image with a convolutional neural network (CNN) into an image category, the computer system comprising:
one or more computer processors, one or more computer-readable storage media, and program instructions stored on the one or more of the computer-readable storage media for execution by at least one of the one or more processors capable of performing a method, the method comprising:
receiving an image with two spatial dimensions for classification by a convolutional neural network (CNN) for classification, the image having an arbitrary rotation in an image;
performing a radial summation preprocessing on the image to generate a radially summed image prior to utilizing the CNN, the radial summation converting the image with two spatial dimensions into a tensor with three spatial dimensions without relying upon a specialized layer of the CNN to perform the radial summation; and
inputting the radially summed image into the CNN to perform an image classification, the image classification classifying the image into an image category; and
classifying the radially summed images into the image category with the CNN.
14. The computer system of claim 13 , wherein the three spatial dimensions further includes an azimuthal angle.
15. The computer system of claim 14 , wherein the radial summation comprises:
performing a weighted sum of radial segments starting from a given pixel of the image for a set number of a plurality of the azimuthal angle.
16. The computer system of claim 13 , wherein the radial summation is performed independently for each image channel.
17. The computer system of claim 13 , wherein the CNN performs three-dimensional convolutions on the radially summed image.