Abstract: A method and system for incorporating regression into a Stacked Auto Encoder utilizing deep learning based regression technique that enables joint learning of parameters for a regression model to train the SAE for a regression problem. The method comprises generating a regression model for the SAE for solving the regression problem, wherein regression model is formulated as a non-convex joint optimization function for an asymmetric SAE. The method further comprises reformulating the non-convex joint optimization function as an Augmented Lagrangian formulation in terms of a plurality of proxy variables and a plurality of hyper parameters. The method comprises splitting the Augmented Lagrangian formulation into sub-problems using Alternating Direction Method of Multipliers and jointly learning parameters for the regression model to train the SAE for the regression problem. The learned weights enable estimating the unknown target values.

Abstract: This disclosure relates generally to a method and system for multi-modal image super-resolution. Conventional methods for multi-modal image super-resolution are performed using joint image based filtering, deep learning and dictionary based approaches which require large datasets for training. Embodiments of the present disclosure provide a joint optimization based transform learning framework wherein a high-resolution (HR) image of target modality is reconstructed from a HR image of guidance modality and a low-resolution (LR) image of target modality. A set of parameters, transforms, coefficients and weight matrices are learnt jointly from a training data which includes a HR image of guidance modality, a LR image of target modality and a HR image of target modality. The learnt set of parameters are used for reconstructing a HR image of target modality. The disclosed joint optimization transform learning framework is used in remote sensing, environment monitoring and so on.

Abstract: Multi-sensor fusion is a technology which effectively utilizes the data from multiple sensors so as to portray a unified picture with improved information and offers significant advantages over existing single sensor-based techniques. This disclosure relates to a method and system for a multi-label classification using a two-stage autoencoder. Herein, the system employs autoencoder based architectures, where either raw sensor data or hand-crafted features extracted from each sensor are used to learn sensor-specific autoencoders. The corresponding latent representations from a plurality of sensors are combined to learn a fusing autoencoder. The latent representation of the fusing autoencoder is used to learn a label consistent classifier for multi-class classification. Further, a joint optimization technique is presented for learning the autoencoders and classifier weights together.

Abstract: This disclosure relates to multi-sensor fusion using Transform Learning (TL) that provides a compact representation of data in many scenarios as compared to Dictionary Learning (DL) and Deep network models that may be computationally intensive and complex. A two-stage approach for better modeling of sensor data is provided, wherein in the first stage, representation of the individual sensor time series is learnt using dedicated transforms and their associated coefficients and in the second stage, all the representations are fused together using a fusing (common) transform and its associated coefficients to effectively capture correlation between the different sensor representations for deriving an inference. The method and system of the present disclosure can find application in areas employing multiple sensors that are mostly heterogeneous in nature.

Abstract: This disclosure relates to systems and methods for solving generic inverse problems by providing a coupled representation architecture using transform learning. Convention solutions are complex, require long training and testing times, reconstruction quality also may not be suitable for all applications. Furthermore, they preclude application to real-time scenarios due to the mentioned inherent lacunae. The methods provided herein require involve very low computational complexity with a need for only three matrix-vector products, and requires very short training and testing times, which makes it applicable for real-time applications. Unlike the conventional learning architectures using inductive approaches, the CASC of the present disclosure can learn directly from the source domain and the number of features in a source domain may not be necessarily equal to the number of features in a target domain.

Abstract: This disclosure relates generally to image processing, and more particularly to method and system for image reconstruction using deep dictionary learning (DDL). The system collects the degraded image as test image and processes the test image to extract sparse features from the test image, at different levels, using dictionaries. The extracted sparse features and data from the dictionaries are used by the system to reconstruct the HR image corresponding to the test image.

Abstract: This disclosure relates generally to image processing, and more particularly to method and system for image reconstruction using deep dictionary learning (DDL). The system collects the degraded image as test image and processes the test image to extract sparse features from the test image, at different levels, using dictionaries. The extracted sparse features and data from the dictionaries are used by the system to reconstruct the HR image corresponding to the test image.

Abstract: This disclosure relates to systems and methods for solving generic inverse problems by providing a coupled representation architecture using transform learning. Convention solutions are complex, require long training and testing times, reconstruction quality also may not be suitable for all applications. Furthermore, they preclude application to real-time scenarios due to the mentioned inherent lacunae. The methods provided herein require involve very low computational complexity with a need for only three matrix-vector products, and requires very short training and testing times, which makes it applicable for real-time applications. Unlike the conventional learning architectures using inductive approaches, the CASC of the present disclosure can learn directly from the source domain and the number of features in a source domain may not be necessarily equal to the number of features in a target domain.

Abstract: A method and system for incorporating regression into a Stacked Auto Encoder utilizing deep learning based regression technique that enables joint learning of parameters for a regression model to train the SAE for a regression problem. The method comprises generating a regression model for the SAE for solving the regression problem, wherein regression model is formulated as a non-convex joint optimization function for an asymmetric SAE. The method further comprises reformulating the non-convex joint optimization function as an Augmented Lagrangian formulation in terms of a plurality of proxy variables and a plurality of hyper parameters. The method comprises splitting the Augmented Lagrangian formulation into sub-problems using Alternating Direction Method of Multipliers and jointly learning parameters for the regression model to train the SAE for the regression problem. The learned weights enable estimating the unknown target values.

Abstract: Motion blur occur when acquiring images and videos with cameras fitted to the high speed motion devices, for example, drones. Distorted images intervene with the mapping of the visual points, hence the pose estimation and tracking may get corrupted. System and method for solving inverse problems using a coupled autoencoder is disclosed. In an embodiment, solving inverse problems, for example, generating a clean sample from an unknown corrupted sample is disclosed. The coupled autoencoder learns the autoencoder weights and coupling map (between source and target) simultaneously. The technique is applicable to any transfer learning problem. The embodiments of the present disclosure implements/proposes a new formulation that recasts deblurring as a transfer learning problem which is solved using the proposed coupled autoencoder.

Abstract: Motion blur occur when acquiring images and videos with cameras fitted to the high speed motion devices, for example, drones. Distorted images intervene with the mapping of the visual points, hence the pose estimation and tracking may get corrupted. System and method for solving inverse problems using a coupled autoencoder is disclosed. In an embodiment, solving inverse problems, for example, generating a clean sample from an unknown corrupted sample is disclosed. The coupled autoencoder learns the autoencoder weights and coupling map (between source and target) simultaneously. The technique is applicable to any transfer learning problem. The embodiments of the present disclosure implements/proposes a new formulation that recasts deblurring as a transfer learning problem which is solved using the proposed coupled autoencoder.