Systems and methods for image processing
Efficient image processing systems and methods for image scene classification and similarity matching are disclosed. The image processing systems encompassed by this disclosure use a deep convolutional neural network to facilitate scene classification by recognizing the context of an image and thereby enabling searches for similar images. These methods and systems are scalable to a large set of images and achieve a higher performance compared to the current state of the art techniques.
1. A method of performing scene classification of an image, the method comprising:
receiving an image into a deep convolutional neural network (DCNN) comprising multiple layers, each layer having multiple nodes, the DCNN having been customized with adjusted weights at each layer;
classifying, by the DCNN, based on the adjusted weights, the image into one or more intermediate output categories, the intermediate output categories being chosen from a set of pre-defined output categories, each of the pre-defined output categories being associated with a separate corresponding binary classifier and a separate corresponding scene classification;
extracting, by the DCNN, based on the adjusted weights, one or more characteristic features of the image; and
further classifying the image using at least the binary classifiers associated with the one or more intermediate output categories, based on the one or more characteristic features of the image, into the scene classifications corresponding to the one or more intermediate output categories or into a category that is not associated with any of the pre-defined output categories.
2. The method of claim 1 wherein determining the one or more intermediate output categories includes computing at least one non-linear transformation of the received image.
3. The method of claim 1 wherein the one or more intermediate output categories determined by the DCNN are one of an indoor category, an outdoor category and a combination thereof.
4. The method of claim 1 wherein the DCNN comprises at least one convolution layer, a pooling layer and a fully connected layer.
5. The method of claim 1 wherein the one or more characteristic features of said image extracted by the DCNN are an output of a first fully connected layer of the DCNN.
6. The method of claim 1 , wherein the image is classified into more than one intermediate output category.
7. The method of claim 1 , wherein the image is classified into only one intermediate output category.
8. A system for scene classification of an image, the system comprising:
a receiver unit for receiving an image;
a base classifier, associated with said receiver unit, wherein the base classifier comprises a deep convolutional neural network (DCNN) comprising multiple layers, each layer having multiple nodes, the DCNN having been customized with adjusted weights at each layer, and wherein the DCNN, based on the adjusted weights, extracts one or more characteristic features of said image and classifies the image into one or more intermediate output categories, the intermediate output categories being chosen from a set of pre-defined output categories, each of the pre-defined output categories being associated with a separate scene classification; and
a set of binary classifiers associated with the base classifier, each binary classifier being associated with a separate pre-defined output category of the set of pre-defined output categories, for further classifying the image, based on the one or more characteristic features of the image, into the scene classifications corresponding to the one or more intermediate output categories or into a category that is not associated with any of the pre-defined output categories.
9. The system of claim 8 wherein the binary classifiers are be a Support Vector Machine.