Patent•
Method for realizing super resolution for image
02 Nov 2016-
TL;DR: In this article, the authors proposed a method for super resolution for an image and belongs to the computer vision field, which includes the following steps of: A1, data preprocessing: a certain number of high-resolution natural images are adopted to form a data set, image blocks are extracted from the data set and Bicubic interpolation downsampling and up-sampling in three times are carried out on the image blocks, and low-resolution images can be obtained; A2, network structure design: a designed convolutional neural network has 4 layers altogether;
Abstract: The invention relates to a method for realizing super resolution for an image and belongs to the computer vision field. The method includes the following steps of: A1, data preprocessing: a certain number of high-resolution natural images are adopted to form a data set, a certain number of image blocks are extracted from the data set, Bicubic interpolation down-sampling and up-sampling in three times are carried out on the image blocks, and low-resolution images can be obtained; A2, network structure design: a designed convolutional neural network has 4 layers altogether; A3, hyper parameter selection: parameters such as network learning rate, learning momentum and batch_size are determined; and A4, network training and super parameter optimization: the convolutional neural network of all images in the training set from low-resolution images to corresponding high-resolution images is trained, and after any one image is inputted into the trained network, a high-resolution image can be obtained, so that the super resolution of the image can be realized.
Citations
More filters
Patent•
31 May 2017
TL;DR: In this paper, a video enhancement and transmission method based on deep learning was proposed, where the size of the video data is greatly reduced through downsampling and video coding, so the video traffic needing to be transmitted is correspondingly reduced and an effect of reducing the bandwidth cost is achieved.
Abstract: The invention discloses a video enhancement and transmission method based on deep learning. Downsampling is carried out on a high-definition video at a video source end, thereby obtaining a low-definition video; the low-definition video is compressed in an existing video coding mode; and the compressed low-definition video is transmitted. The size of the video data is greatly reduced through downsampling and video coding, so the video traffic needing to be transmitted is correspondingly reduced, and an effect of reducing the bandwidth cost is achieved. At a user receiving end, a user receives the low-definition video, reconstructs the low-definition video by employing a super-resolution image reconstruction method of the deep learning and restores the low-definition video into a high-resolution video for the user to watch, so the video transmission bandwidth cost is effectively reduced. According to the method, the video transmission bandwidth cost is reduced by at least 50%; the resolution of the live broadcast video is improved; and the watching experience of a user is improved.
25 citations
Patent•
29 Dec 2017
TL;DR: In this article, a multi-licence-plate sharpness method based on search and an apparatus thereof is presented. But the method is limited to the acquisition of a target image including a target vehicle, according to characteristic information of the target vehicle.
Abstract: An embodiment of the invention provides a multi-license-plate sharpness method based on search and an apparatus thereof. The method comprises the following steps of acquiring a target image including a target vehicle; according to characteristic information of the target vehicle, acquiring a plurality of vehicle images including the target vehicles; from the plurality of vehicle images, extracting license plate images corresponding to the plurality of vehicle images; based on a single image generation network, carrying out sharpness processing on the extracted license plate images and acquiring clear license plate images corresponding to the extracted license plate images; from the acquired clear license plate images, selecting a first preset quantity of the clear license plate images; and based on a multi-image generation network, synthesizing the selected clear license plate images into one super-resolution license plate image. In the technical scheme, sharpness is performed on the plurality of vehicle images and the images are synthesized into one image; and information complementation among the plurality of license plates is fully used so that one clear license plate image with complete information can be acquired.
10 citations
Patent•
20 Oct 2017
TL;DR: In this paper, a cGAN algorithm-based super-resolution image recovery method was proposed, which is capable of training three-channel colored images to achieve better recovery effect, and the key points of the method comprises that (1) the training time is short, (2) 3-channel coloured maps can be directly trained, (3) much image preprocessing is not required, and (4) the models obtained through training can be used once for all.
Abstract: The invention discloses a cGAN algorithm-based super-resolution image recovery technology, and relates to super-resolution recovery for deep convolutional antagonistic neural network images. The existing method needs to carry out a lot of training, the network layers are relatively shallow and the feature weight values included in models obtained through the training are not complete enough, so that the final recovery effect is not ideal. Greyscale maps are used during the training, so that the super-resolution recovery effect for three-channel colored images is not good. Aiming at the defects in the prior art, the invention provides a cGAN algorithm-based super-resolution image recovery method which is capable of training the colored images to achieve better recovery effect. The key points of the method comprises that (1) the training time is short, (2) three-channel colored maps can be directly trained, (3) much image preprocessing is not required, and (4) the models obtained through training can be used once for all.
9 citations
Patent•
22 Mar 2017
TL;DR: In this article, an improved deep learning-based intelligent camera image blind super-resolution system was proposed, which consists of: step one, statistical testing to obtain a common fuzzy kernel in photo shooting; step two, establishing a training database, carrying out corresponding fuzzy quality degradation on the database, and constructing a clear quality degradation image pair; step three, training by using the fuzzy kernel and a low-resolution image block as inputs and a high-resolution block as an output.
Abstract: The invention discloses an improved-deep-learning-based intelligent camera image blind super-resolution system. The system comprises: step one, carrying out statistic testing to obtain a common fuzzy kernel in photo shooting; step two, establishing a training database, carrying out corresponding fuzzy quality degradation on the database, carrying out several kinds of typical downsampling, and constructing a clear quality degradation image pair; step three, carrying out network training by using the fuzzy kernel and a low-resolution image block as inputs and a high-resolution block as an output; step four, guiding a neural network model network into an intelligent camera shooting system to form model data known in advance; and step five, before and after shooting by a user, a restoration model is used for carrying out automatic restoration on a shot photo after an image blind super-resolution function is started. Using the method, the resolution of photo shooting by the intelligent camera can be improved on the condition of hardware costs do not increase.
7 citations
Patent•
08 Mar 2017
TL;DR: In this article, a high-quality image magnification method, which comprises an off-line learning method and an online processing method, is described, and the edge of an image through the network is clear and sharp without artificial effects such as serrations, certain noise is added into the training samples, so that the network has a capability of removing encoding noise.
Abstract: The invention discloses a high-quality image magnification method, which comprises an off-line learning method and an online processing method. The invention further discloses a high-quality image method system. According to the high-quality image magnification method and the high-quality image method system, a deep learning technology based on a convolution neural network is adopted, a non-linear mapping relation from blur edge to clear edge can be obtained through learning of a large number of samples, the edge of an image through the network is clear and sharp without artificial effects such as serrations, certain noise is added into the training samples, so that the network has a capability of removing encoding noise, and finally the image is subjected to edge sharpening by adopting an image enhancement algorithm without affecting a flat region, therefore, the noise of the flat region cannot be increased while enhancing edge sharpness, and the user can have a strong subjective feeling.
3 citations
References
More filters
06 Sep 2014
TL;DR: This work proposes a deep learning method for single image super-resolution (SR) that directly learns an end-to-end mapping between the low/high-resolution images and shows that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network.
Abstract: We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) [15] that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage.
4,445 citations
Patent•
18 Nov 2015
TL;DR: In this article, an image super-resolution reconstruction method based on a deep belief network is proposed for the image processing, which consists of obtaining a low-resolution image, performing interpolation amplification on the low resolution image to the needed size, using a repeated partitioning sampling method to obtain a low resolution brightness image block, inputting the low- resolution image block and using the deep belief networks which is trained in advance to predict the high resolution image blocks, performing neighborhood regularization optimization solution on an obtained fitting result.
Abstract: The invention discloses an image super resolution reconstruction method based on a deep belief network, relating to the image processing. The image super resolution reconstruction method based on the deep belief network comprises steps of obtaining a low resolution image, performing interpolation amplification on the low resolution image to the needed size, using a repeated partitioning sampling method to obtain a low resolution brightness image block, inputting the low resolution image block, using the deep belief network which is trained in advance to predict the high resolution image block, performing neighborhood regularization optimization solution on an obtained fitting result, combining all high resolution brightness image blocks to obtain a high resolution and brightness image, combining the high resolution and brightness image with the values of the other two channels which are obtained in advance, and converting high resolution and brightness image to the image expressed by color RGB to obtain a predicted high resolution image. The invention realizes the super-resolution reconstruction of the single-frame image, improves the peak signal-to-noise ratio, obtains a clear rim and a rich texture of a constructed image, and can be used for the video safety monitoring, medical digital imaging and spaceflight detection.
20 citations
Patent•
18 Nov 2015
TL;DR: In this paper, a bilateral-circulation convolution network-based video super-resolution method is proposed, which comprises the steps as follows: establishing bilateralcirculation networks, comprising a forward circulation subnetwork and a backward circulation sub-network according to time sequence, wherein each circulation sub network comprises an input sequence layer, two implication sequence layers and an output sequence layer from the bottom to the top, and each sequence layer comprises a plurality of states corresponding to video frames in different moments.
Abstract: The invention discloses a bilateral-circulation convolution network-based video super-resolution method. The method comprises the steps as follows: establishing bilateral-circulation networks, comprising a forward circulation sub-network and a backward circulation sub-network according to time sequence, wherein each circulation sub-network comprises an input sequence layer, two implication sequence layers and an output sequence layer from the bottom to the top, and each sequence layer comprises a plurality of states corresponding to video frames in different moments; connecting these states by using three convolution operations, comprising a feed-forward convolution, a cyclic convolution and a condition convolution so as to obtain a bilateral-circulation convolution network; transmitting a trained video in the established bilateral-circulation convolution network, and using a stochastic gradient descent algorithm to minimize mean square error between a predicated high resolution video and an actual high resolution video so as to iteratively optimize the weight of the network, and obtain the final bilateral-circulation convolution network; and inputting low-resolution video sequence to be processed in a final bilateral-circulation convolution network mode to obtain a corresponding super-resolution result.
7 citations