System and method for the fusion of bottom-up whole-image features and top-down enttiy classification for accurate image/video scene classification
Described is a system and method for accurate image and/or video scene classification. More specifically, described is a system that makes use of a specialized convolutional-neural network (hereafter CNN) based technique for the fusion of bottom-up whole-image features and top-down entity classification. When the two parallel and independent processing paths are fused, the system provides an accurate classification of the scene as depicted in the image or video.
1. A system for scene classification, the system comprising:
one or more processors and a memory, the memory being a non-transitory computer-readable medium having executable instructions encoded thereon, such that upon execution of the instructions, the one or more processors perform operations of:
operating at least two parallel, independent processing pipelines on a whole image to generate independent results, wherein the at least two parallel, independent processing pipelines includes both an entity processing pipeline and a whole image processing pipeline, wherein the entity processing pipeline operates on the whole image and uses a convolutional neural network (CNN) which scans the whole image to identify a number and type of entities in the whole image, resulting in an entity feature space, and wherein the whole image processing pipeline uses a CNN to extract visual features from the whole image, resulting in a visual feature space; and
fusing the independent results of the entity and whole image processing pipelines to generate a fused scene class, such that in fusing the independent results to generate the fused scene class, two classifiers are trained separately for each of the visual feature space and entity feature space to generate independent class probability distributions over scene types, with the independent class probability distributions being multiplied and renormalized to generate the fused scene class; and
electronically controlling machine behavior based on the fused scene class of the whole image.
2. The system as set forth in claim 1 , wherein the entity processing pipeline identifies and segments potential object locations within the image or video and assigns a class label to each identified and segmented potential object within the image or video.
3. The system as set forth in claim 1 , wherein the entity feature space includes a bag of words histogram feature.
4. The system as set forth in claim 1 , wherein electronically controlling machine behavior includes at least one of labeling data associated with the image or video with the fused scene class, displaying the fused scene class with the image or video, controlling vehicle performance, or controlling processor performance.
5. The system as set forth in claim 1 , further comprising an operation of displaying the image or video with a label that includes the fused scene class.
6. A computer program product for scene classification, the computer program product comprising:
a non-transitory computer-readable medium having executable instructions encoded thereon, such that upon execution of the instructions by one or more processors, the one or more processors perform operations of:
operating at least two parallel, independent processing pipelines on a whole image to generate independent results, wherein the at least two parallel, independent processing pipelines includes both an entity processing pipeline and a whole image processing pipeline, wherein the entity processing pipeline operates on the whole image and uses a convolutional neural network (CNN) which scans the whole image to identify a number and type of entities in the whole image, resulting in an entity feature space, and wherein the whole image processing pipeline uses a CNN to extract visual features from the whole image, resulting in a visual feature space; and
fusing the independent results of the entity and whole image processing pipelines to generate a fused scene class, such that in fusing the independent results to generate the fused scene class, two classifiers are trained separately for each of the visual feature space and entity feature space to generate independent class probability distributions over scene types, with the independent class probability distributions being multiplied and renormalized to generate the fused scene class; and
electronically controlling machine behavior based on the fused scene class of the whole image.
7. The computer program product as set forth in claim 6 , wherein the entity processing pipeline identifies and segments potential object locations within the image or video and assigns a class label to each identified and segmented potential object within the image or video.
8. The computer program product as set forth in claim 6 , wherein the entity feature space includes a bag of words histogram feature.
9. The computer program product as set forth in claim 6 , wherein electronically controlling machine behavior includes at least one of labeling data associated with the image or video with the fused scene class, displaying the fused scene class with the image or video, controlling vehicle performance, or controlling processor performance.
10. The computer program product as set forth in claim 6 , further comprising an operation of displaying the image or video with a label that includes the fused scene class.
11. A computer implemented method for scene classification, the method comprising an act of:
causing one or more processers to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:
operating at least two parallel, independent processing pipelines on a whole image to generate independent results, wherein the at least two parallel, independent processing pipelines includes both an entity processing pipeline and a whole image processing pipeline, wherein the entity processing pipeline operates on the whole image and uses a convolutional neural network (CNN) which scans the whole image to identify a number and type of entities in the whole image, resulting in an entity feature space, and wherein the whole image processing pipeline uses a CNN to extract visual features from the whole image, resulting in a visual feature space; and
fusing the independent results of the entity and whole image processing pipelines to generate a fused scene class, such that in fusing the independent results to generate the fused scene class, two classifiers are trained separately for each of the visual feature space and entity feature space to generate independent class probability distributions over scene types, with the independent class probability distributions being multiplied and renormalized to generate the fused scene class; and
electronically controlling machine behavior based on the fused scene class of the whole image.
12. The method as set forth in claim 11 , wherein the entity processing pipeline identifies and segments potential object locations within the image or video and assigns a class label to each identified and segmented potential object within the image or video.
13. The method as set forth in claim 11 , wherein the entity feature space includes a bag of words histogram feature.
14. The method as set forth in claim 11 , wherein electronically controlling machine behavior includes at least one of labeling data associated with the image or video with the fused scene class, displaying the fused scene class with the image or video, controlling vehicle performance, or controlling processor performance.
15. The method as set forth in claim 11 , further comprising an operation of displaying the image or video with a label that includes the fused scene class.