Abstract: A neural network model training apparatus for enhancing image detail is provided. The apparatus includes a memory and at least one processor configured to obtain a low quality input image patch and a high quality input image patch, obtain a low quality output image patch by inputting the low quality input image patch to a first neural network model, obtain a high quality output image patch by inputting the high quality input image patch to a second neural network model, and train the first neural network model based on a loss function set to reduce a difference between the low quality output image patch and the high quality input image patch, and a difference between the high quality output image patch and the high quality input image patch. The second neural network model is identical to the first neural network model.

Abstract: A method for training a convolutional neural network to reconstruct an image. The method includes forming a common loss function basing on the left and right images (IL, IR), reconstructed left and right images (I?L, I?R), disparity maps (dL, dR), reconstructed disparity maps (d?L, d?R) for the left and right images (IL, IR) and the auxiliary images (I?L, I?R) and training the neural network based on the formed loss function.

Abstract: An image processing device includes an encoder configured to receive a blurry image and generate a global feature map of the image, a merging unit configured to merge the global feature map and blur information, a decoder configured to generate a feature tensor and weight tensors, a recurrent refinement module configured to perform recurrent feature filtering, and an image reconstruction module configured to reconstruct a deblurred image where the image processing device is configured to estimate an image global shift and to activate or deactivate the recurrent refinement module based on the estimation.

Abstract: A method for training a convolutional neural network to reconstruct an image. The method includes forming a common loss function basing on the left and right images (IL, IR), reconstructed left and right images (I?L, I?R), disparity maps (dL, dR), reconstructed disparity maps ((d?L, d?R) for the left and right images (IL, IR) and the auxiliary images (I?L, I?R) and training the neural network based on the formed loss function.

Abstract: A neural network model training apparatus for enhancing image detail is provided. The apparatus includes a memory and at least one processor configured to obtain a low quality input image patch and a high quality input image patch, obtain a low quality output image patch by inputting the low quality input image patch to a first neural network model, obtain a high quality output image patch by inputting the high quality input image patch to a second neural network model, and train the first neural network model based on a loss function set to reduce a difference between the low quality output image patch and the high quality input image patch, and a difference between the high quality output image patch and the high quality input image patch. The second neural network model is identical to the first neural network model.

Abstract: A method for training a convolutional neural network to reconstruct an image. The method includes forming a common loss function basing on the left and right images (IL, IR), reconstructed left and right images (I?L, I?R), disparity maps (dL, dR), reconstructed disparity maps (d?L, d?R) for the left and right images (IL, IR) and the auxiliary images (I?L, I?R) and training the neural network based on the formed loss function.

Abstract: A method and apparatus for separating a text and figure of a document image are provided. The method of separating the text and the figure of the document image includes acquiring a document image, dividing the document image into a plurality of regions of interest, acquiring a feature vector by using a two-dimensional (2D) histogram by resizing the regions of interest and extracting a connection component of the regions of interest, acquiring a transformation vector of the feature vector by using a kernel, obtaining a cluster center of the transformation vector, and performing clustering on the cluster center to acquire a supercluster, and classifying the supercluster into one of a text class and a figure class, based on the number of superclusters.

Abstract: A method for training a convolutional neural network to reconstruct an image. The method includes forming a common loss function basing on the left and right images (IL, IR), reconstructed left and right images (I?L, I?R), disparity maps (dL, dR), reconstructed disparity maps (d?L, d?R) for the left and right images (IL, IR) and the auxiliary images (I?L, I?R) and training the neural network based on the formed loss function.

Abstract: A motion estimation method for video data including frames and an apparatus therefor are provided. The motion estimation method includes determining whether a current frame unit for which motion estimation is to be performed corresponds to a double block, when the current frame unit corresponds to a double block, acquiring a candidate vector set corresponding to a first single block included in the double block, as a candidate vector set of the double block, individually calculating a confidence function value of each candidate vector included in the candidate vector set of the double block, for the first single block and a second single block included in the double block, and acquiring an estimated motion vector of the first single block and an estimated motion vector of the second single block, based on the calculated confidence function value of each candidate vector.

Abstract: The invention relates to a method for protecting a video frame sequence against random and/or burst packet losses, the method comprising: partitioning the video frame sequence into a plurality of interleaved video frame subsequences; independently encoding the plurality of interleaved video frame subsequences by a differential video codec; generating for at least one of the plurality of interleaved video frame subsequences at least one B frame predicted from another one of the plurality of interleaved video frame subsequences; and using the at least one B frame as redundant B frame for protecting the video frame sequence.

Abstract: A motion estimation method for video data including frames and an apparatus therefor are provided. The motion estimation method includes determining whether a current frame unit for which motion estimation is to be performed corresponds to a double block, when the current frame unit corresponds to a double block, acquiring a candidate vector set corresponding to a first single block included in the double block, as a candidate vector set of the double block, individually calculating a confidence function value of each candidate vector included in the candidate vector set of the double block, for the first single block and a second single block included in the double block, and acquiring an estimated motion vector of the first single block and an estimated motion vector of the second single block, based on the calculated confidence function value of each candidate vector.

Abstract: A method and apparatus for separating a text and figure of a document image are provided. The method of separating the text and the figure of the document image includes acquiring a document image, dividing the document image into a plurality of regions of interest, acquiring a feature vector by using a two-dimensional (2D) histogram by resizing the regions of interest and extracting a connection component of the regions of interest, acquiring a transformation vector of the feature vector by using a kernel, obtaining a cluster center of the transformation vector, and performing clustering on the cluster center to acquire a supercluster, and classifying the supercluster into one of a text class and a figure class, based on the number of superclusters.

Abstract: The invention relates to a method for protecting a video frame sequence against random and/or burst packet losses, the method comprising: partitioning the video frame sequence into a plurality of interleaved video frame subsequences; independently encoding the plurality of interleaved video frame subsequences by a differential video codec; generating for at least one of the plurality of interleaved video frame subsequences at least one B frame predicted from another one of the plurality of interleaved video frame subsequences; and using the at least one B frame as redundant B frame for protecting the video frame sequence.