152 results about "Rain removal" patented technology

The invention provides a single image rain removal method based on a convolutional neural network and relates to image processing. The method includes the following steps that firstly, manual rain addition is conducted on clean and clear images through a screen blend model to form corresponding rain images, and an image library is built; secondly, the system structure of the convolutional neural network is determined; thirdly, corresponding rain image blocks and rain-free image blocks with the size being 64*64 are obtained from the first step and serve as training samples for training; fourthly, the single image blocks are obtained in an overlapped mode and input in a trained rain removal filter system to obtain the corresponding rain-free image blocks, and weighted averaging is conducted on the image blocks to obtain rain-free images. The method solves the problem that a single image rain removal method based on dictionary learning is long in consumed time, achieves rain removal and guarantees definition of background images, the rain-free images can be quickly obtained after the rain images are input, and the requirement of embedded equipment for real-time processing is met.

The invention discloses a single image fine rain removal method based on a depth convolutional neural network. Firstly, through carrying out background texture structure extraction, nonlinear mapping and rain line area restoration on an inputted rain graph by an initial rain removal network, an initial clear rain-free image is finally obtained; the initial clear rain-free image and the original image are inputted to a fine rain removal network with a single convolutional layer at the same time, more details in the background area are thus restored, and a high-definition rain removal image is finally obtained. Through adopting a caffe framework, the initial rain removal network and the fine rain removal network are trained, parameters of each convolutional layer are obtained precisely, fine rain removal processing is carried out on a rain image, and compared with the traditional convolutional neural network rain removal method, the method of the invention can obtain a higher-quality rain-free image, the practicability is strong, and the method can be widely applied to more scenes.

The invention provides a wind blowing and precipitation system, which is used for airplane windscreen rain removal system tests. The wind blowing and precipitation system comprises a wind blowing system, a precipitation system and a measurement and control system, wherein the wind blowing system adopts a straight-flow low-speed wind tunnel and a variable-frequency fan is arranged in the wind tunnel; the precipitation system comprises a water pump and a precipitation device, the water pump delivers tap water to the precipitation device, the precipitation device comprises a frame and a wing-shaped plate installed in the frame, a plurality of nozzles are arranged at the rear edge of the wing-shaped plate, and the spray direction of the plurality of nozzles is in parallel with the chord length direction of the wing-shaped plate; the measurement and control system comprises a wind speed sensor, a rainfall sensor and a controller, wherein the controller adjusts the frequency of the variable-frequency fan to adjust wind speed according to the feedback of the wind speed sensor and adjusts the revolving speed of the water pump to adjust rainfall according to the feedback of the rainfall sensor. The invention further provides a rainfall simulation method using the wind blowing and precipitation system.

The invention relates to a refined single image rain removing method based on a generative adversarial network. Inputting the rain image into a rain streak estimation network; obtaining an estimated rain streak graph; connecting with an input image to form a multi-channel image; the rain-free image generated by the generation model is input into the discriminator for judgment, the generator is optimized according to a judgment result, a generator network with high rain removal capability is finally obtained, the output of the generation model serves as the input of the image refinement network, the image is further processed, and a final rain-free image is obtained. The algorithm provided by the invention is an end-to-end algorithm, and any additional preprocessing and post-processing arenot needed. Compared with other work of using a generative adversarial network to carry out a single image area, two auxiliary networks are provided, and the image rain removal effect can be further improved while the calculated amount is not significantly increased.

The invention discloses an image rain removal method based on a residual adversarial refinement network, which can effectively remove raindrops in an image and restore a real scene image. According tothe method, rain removal is divided into two stages, in the first stage, a residual image is removed from an original image, a clear image is recovered, the recovered image is input into the second stage, background information of the part shielded by raindrops is recovered, the image is refined, and the quality is improved. According to the method, a rain image is regarded as synthesis of a clear image and a residual image, based on the principle, the residual image is obtained from an original rain image through a residual network, and the residual image is subtracted from the original image to obtain an output image of a first stage; in the second stage, the generative adversarial network is adopted to take the output image and the residual image in the first stage as input, the residual image is taken as an attention mechanism, and the network is assisted to recover a more real image. The invention further provides an image rain removing system.

The invention discloses a single picture bidirectional rain removing method based on picture block raindrop density. The method can remove rain from a picture with non-uniform raindrop distribution. The method remove rain of picture blocks by different degrees, according to raindrop intensities of different areas of a single picture predicted by a classifier, and excessive rain removal and incomplete rain removal are prevented. A plurality of picture block bidirectional rain removing modules are trained to remove rain from picture blocks with different sizes, and reinforcing the relationship between the encoder and the generator through the generated hidden layer codes to ensure that the generator can remove rain according to the raindrop intensity. A repairing module is used for repairingthe spliced image subjected to rain removal according to the unsmooth and distorted phenomena. Considering that dense raindrops generate an atomization phenomenon in a rainy day picture, a defoggingmodule is added to a defogging network. Rain removal is carried out on pictures which are distributed unevenly in rainy days, heavy rain pictures are effectively removed through a circulating structure, and meanwhile the atomization phenomenon generated in the rainy days is removed.

The invention relates to the technical field of image processing, in particular to an image rain removing method and device, a readable storage medium and terminal equipment. The method provided by the invention comprises the steps of obtaining an image with rain; inputting the rain-carrying image into the trained generative adversarial network model to obtain a clear rain-removed image of the rain-carrying image output by the generative adversarial network model; wherein the generative adversarial network model comprises a generative model and a discriminant model which are set as adversarialtraining; wherein the generation model is a full convolutional network model obtained by carrying out feature supervision training on the first image by utilizing features extracted from the second image; wherein the first image is a first training rain image; wherein the second image is a first training clear image corresponding to the first training rain image, so as to train the generative adversarial network model through feature supervision of the clear image, thereby optimizing the feature extraction function of the generative adversarial network model, improving the extraction accuracyof the original texture features of the image, and improving the rain removal effect of the image.

A method for removing rain in a video based on noise modeling is disclosed. Under the assumption of a low-rank background, the rain bar noise component and the moving foreground in the video are simultaneously estimated. First, video data containing rain noise is acquired and a model is initialized; a rain map generation model is created according to the characteristics of the rain noise and the video foreground; the structural characteristics of the rain imaging in the video-a rain bar formed by moving rain droplets on each small block in an image is identical in the direction, the small block prior distribution of the rain bar is established; a moving object detection model is established according to the characteristics of the video foreground sparsity; the model is converted into a rain removal model under the maximum likelihood estimation framework; a rain-containing video and the rain removal model are applied to get a rain-removed video and other statistical variables, and the rain-removed video is output. The method aims to build a high-quality video rain removal model based on a rain map generation principle and rain bar noise structure characteristics, thereby more accurately allowing the video rain removal technology to be widely applied to complex raining scenes with the moving foreground.

Systems and methods for providing rain and ice removal on a windshield are. provided. The system may be implemented on a vehicle, such as an aircraft. In one embodiment, a system includes an injection molded windshield and a frame coupled to the injection-molded windshield and to the vehicle. The frame includes a channel that directs at least one of air or fluid onto an exterior surface of the windshield. The frame includes one or more one-way check valves. A fluid pump pumps fluid through the channel and onto the windshield. A reservoir stores de-icing fluid that is retrieved by the fluid pump. Air sources pump air through the channel and onto the one or more windshields. A controlling device controls the fluid pump or the air sources.

The invention relates to a single-image rain removing method of a multi-channel multi-scale convolutional neural network. A multi-channel multi-scale convolutional neural network is established and used for extracting feature information of a rainy image and mapping the feature information to the direction of a rainless image. Firstly, a rain image is decomposed through guided filtering, and a low-frequency background image and a higher-frequency rain line image are obtained; and then the low-frequency background image and the higher-frequency rain line image are sent to a convolutional neuralnetwork for feature learning to obtain a rain line-removed low-frequency image and a rain line-removed higher-frequency image, and finally the rain line-removed low-frequency image and the rain line-removed higher-frequency image are fused to obtain a rain line-removed image. In the convolutional neural network, a multi-scale feature map is extracted from a low-frequency background image and a higher-frequency rain line image. And meanwhile, when the network model is constructed, the multi-scale feature information of the image is extracted by using hole convolution instead of common convolution, so that richer image features are obtained, and the rain removal performance of the algorithm is improved. Experiments prove that the rain line in the image can be effectively removed, and the rain-removed image is clear and high in detail retention degree.

The invention discloses a single image rain removal method based on a composite residual network and deep supervision, and the method comprises the following steps: constructing a training set, and collecting images with rain in a plurality of directions in a natural scene and corresponding clean images; preprocessing: randomly selecting an image pair from the training set as the input of the network; extracting features, and inputting the image blocks with rain into a composite residual error network containing a plurality of residual error modules for processing to obtain multi-level features; performing image reconstruction: splicing the output features of each residual module, inputting the spliced output features into a convolution layer to obtain a three-channel image, and taking thethree-channel image as a final restored image; and supervising the output of each residual error module by using the clean image, i.e., deeply supervising, so as to optimize network parameters. According to the method, the rain strips in multiple directions are effectively removed, scene detail information can be well reserved, and the method can be applied to various image restoration applications.

The invention relates to an image rain removal method and system. The method comprises the steps: S1, constructing an image training database, wherein the image training database comprises a pluralityof no rain-rain-only rain grain image pairs; S2, according to the no rain-rain-only rain grain image pairs in the image training database, constructing a twin convolution network structure for rain removal; S3, performing filtering on an image to be subjected to rain removal to obtain high frequency information and low frequency information of the image to be subjected to rain removal; and S4, inputting the high frequency information of the image to be subjected to rain removal into the twin convolution network structure for rain removal to obtain a high frequency information of a corresponding no-rain image, and adding the low frequency information of a rain image into the obtained high frequency information of a no-rain image to obtain the corresponding no-rain image. According to the method, the twin convolution network structure is constructed by the no rain-rain-only rain grain image pairs, operation is simplified, the construction processing speed is high, the instantaneity is good, and the clear no-rain image can be obtained by constructing the twin convolution network structure with high robustness.

The embodiment of the invention provides a synchronous rain and fog synthesis and removal method and a device in an image. The synthesis method comprises steps: any real image with no rain and no fog is selected as a reference image; and the reference image is fused through a rain image model RRM, and an image with a rain and a fog close to a real scene is synthesized. The removal method comprises steps: multiple images with the rain and the fog synthesized are selected as a training set; a totally convolutional neural network is trained, and a totally convolutional neural network after training is acquired; and a photographed real image with rain is acquired, the photographed real image with rain is inputted to the totally convolutional neural network after training, and an image after rain removal is outputted. The method realizes problems of synchronous rain and fog synthesis and removal in the image.

The invention relates to a video image rain removal method and system. The video image rain removal method comprises: step a: inputting a video frame image, and converting the video frame image from an RGB color space to a YCbCr color space; step b: obtaining a Y component sequence of an image video frame, and segmenting an image pixel in a Y component by utilizing a fast fuzzy C-means cluster algorithm to obtain a rain region image; and step c: removing raindrops in the rain region image by utilizing an alpha mixing technology to obtain a final rain-free image. According to the video image rain removal method and system, the time required for raindrop removal is greatly shortened through color space conversion, so that the rain removal efficiency is improved; optimal fuzzy C-means classification of pixels is obtained by utilizing the fast fuzzy C-means cluster algorithm, so that the accuracy of the algorithm is improved; raindrop false detection elimination is performed by adopting the photometric property of the raindrops, so that the identification of dynamic scene raindrops is enhanced; and raindrop removal is performed by utilizing the alpha mixing technology, so that it is ensured that sharp edges are not left after the raindrops are removed.

The invention provides an image rain removal method based on an attention mechanism and a gating circulation unit, and the method comprises the steps: firstly constructing an image rain removal network architecture based on the attention mechanism and the gating circulation unit, wherein the image rain removal network architecture includes 6 modules, each of the first five modules comprises a gating circulation unit, a spatial attention module and an activation function; wherein the module 1 is used as an encoder, expansion convolution with expansion factors of 1, 2, 4 and 8 is respectively used in the module 2 to the module 5, and the sizes of corresponding receiving domains are respectively 5 * 5, 9 * 9, 17 * 17 and 33 * 33; the module 6 comprises a convolution layer, a channel attentionmodule and an activation function, and a 1 * 1 convolution layer is connected behind the module 6 and serves as a decoder to generate residual error mapping; then, selecting the MSE and the SSIM as loss functions; and finally, training the constructed network architecture. According to the method, the problems of rain stripe residues and detail blurring occurring in processing of the image containing the dense rain stripes are solved, the rain lines in the image are removed, meanwhile, the detail parts of the image are reserved, and the image definition is greatly improved.

The invention belongs to the technical field of image rain removal, and provides a single-image-oriented rain removal method based on a cascaded hole convolutional neural network, which is used for solving the problem of restoration of a single image shot in rainy days. The method comprises the following steps: firstly, modeling rainwater, and dividing a rain image into a rainwater region layer, arainwater layer and a background layer; extracting a rainwater region layer image from an input image through cascaded multi-channel convolutional neural networks with different void ratios, obtaining a rainwater layer image through convolution, and obtaining a rain-removed background layer image through convolution and summation of the input image. Details of different scales of the image are effectively extracted through the cascaded hole convolutional neural network, the network adopts a residual network structure to increase the network depth, and the over-fitting problem is avoided; an evaluation experiment is carried out on a public data set, and the experiment shows that compared with a single-image rain removal classic method, the peak signal-to-noise ratio (PSNR) can be improvedby 2-8, and the image structural similarity (SSIM) can be improved by 0.04-0.22.

The invention discloses a single image rain removing method based on an optimization algorithm in combination with a residual network, and belongs to the technical field of computer vision application. The method comprises: using an alternating direction multiplier method ADMM for solving a rainy day image imaging model, embedding a residual error network and a noise reduction algorithm into an ADMM framework to serve as background priori and rain priori for iteration, and dividing an image shot in rainy days into a rain-free clear background part and a rain trace part. In the residual network, a synthetic rain map/clear background image pair is used as a training set for training to describe image background priori. Experimental verification proves that the PSNR value of the rain removalresult of the ADMM algorithm embedded with the residual error network is higher than that of other rain removal algorithms. In addition, other existing rain removing algorithms are embedded into the ADMM rain removing algorithm to serve as background prior iteration, and the obtained rain removing effect is superior to the effect of an original algorithm.

The invention relates to a single image rain removing method based on a compressed reward and punishment neural network reusing original information. The single image rain removing method comprises the following steps: firstly, decomposing a rain image into a low-frequency image layer and a high-frequency image layer by utilizing rapid guide filtering; inputting the high-frequency image layer intoa neural network which combines a compressed reward and punishment neural network structure block, batch normalization processing and an original information connection reuse mode provided by the method to carry out feature learning and extraction, and removing rain lines in the neural network; and finally, adding the high-frequency layer without the rain line to the original low-frequency layerto obtain a final rain removal result. According to the single image rain removing method, rain removal is carried out on a single rain image, and compared with an existing traditional rain removal method and a rain removal method based on deep learning, a rain-free image with higher quality can be obtained; and in addition, the network proposed by the method is based on the compressed reward andpunishment neural network, and the compressed reward and punishment structure block used by the network proposed by the method can well describe the relationship between the feature channels, therebyimproving the expression ability of the network and improving the rain removal effect.

A context aggregation residual single image rain removal method based on a convolutional neural network comprises the following specific steps: step 1, preprocessing an image; 2, constructing a convolutional neural network model; 3, training the convolutional neural network model obtained in the step 2 by adopting the preprocessed image obtained in the step 1 to obtain a rain removal convolutionalneural network model, constraining the rain removal convolutional neural network model by utilizing a loss function, and then performing back propagation to update parameters to obtain a trained rainremoval network model; and 4, inputting an image to be processed with rain into the trained rain removal network model, and finally outputting the rain-removed image. Due to the fact that the convolutional neural network model is constructed, more detailed features are obtained, more details are obtained, implementation is easy, and the image rain removing effect is good.

The invention provides an image rain removing method and a device, comprising: an image to be detected with raindrops is separated into a high-frequency component image and a low-frequency component image; the high-frequency component image is input to the trained convolution neural network with residual structure, and the rain-free image is outputted; the rain-free image is synthesized with the image to be detected, and the rain-free image is obtained; the rain-free image is inputted to the trained convolution neural network with discriminant structure, and the category of the rain-free imageand the raindrop-free image corresponding to the image to be detected are outputted; the category is inputted to the convolution neural network with a residual structure, and the parameters of the convolution neural network with residual structure are updated to obtain the final rain-free image. The final rain-free image obtained by the invention retains the texture details of the rain-free areain the image to be detected, so that the final rain-free image is close to the rain-free image corresponding to the image to be detected, and the important factors in the image to be detected are wellretained.

The invention belongs to the technical field of computer vision, and provides a time-domain-based rain line decomposition and space structure guided video rain removal method, being characterized in that rain stripes of each frame are processed from two aspects of position and intensity, and a learnable decomposition mode is defined to learn rain distribution, and a position guide graph acts on asingle-frame rain removing block related to intensity to accurately remove the rain stripes; and secondly, a multi-frame fusion module with an edge guide graph is constructed to fuse time-space information, recover a background and retain more details at the same time. A large number of experiments show that compared with other latest methods, the algorithm is excellent in video rain removal task.Ablation experiments about a network architecture fully show that our network is valid.

The invention belongs to the technical field of image processing, and particularly relates to an image rain removal method and system. The image rain removal method comprises: step a: inputting a video frame image and converting the video frame image from an RGB color space to an HSL color space; step b: extracting H and S channel parameters of the HSL color space; detecting edges of a moving object by using the H and S channel parameters and screening pixel points polluted by rain drops; and step c: removing the pixel points polluted by the rain drops by using an an anisotropic diffusion edge protection algorithm to obtain a rain-removed image. With the adoption of the image rain removal method and system provided by the embodiments of the invention, the video frame image is converted from the RGB color space to the HSL color space, so that time complexity of the algorithm is reduced, and an application range and instantaneity of the algorithm are improved, therefore, erroneous determination of the rain drops and the moving object is avoided; and the rain drops are removed by using the anisotropic diffusion edge protection algorithm to obtain a better rain removing effect, and accuracy and robustness of the algorithm are improved.

The invention relates to an image rain removing method and system based on a static rain pattern, comprising the following steps of: constructing an image training database; according to the rain-purerain pattern in the image training database, constructing the static rain pattern detection network structure; filtering the image for rain removal to obtain high-frequency information and low-frequency information of the image to be rained; inputting the high frequency information of the image for rain removel into the static rain pattern detection network structure to obtain the high frequencyinformation of the rain pattern; Subtracting the high-frequency information of the rain pattern from the high-frequency information of the rainless image to obtain the high-frequency information of the rainless image; adding the high-frequency information of the rain-free image and the low-frequency information of the image for rain removal to obtain the rain-free image. The invention improves the detection capability of the static rain ripple detection network structure by removing the low-frequency information and other interference network function items. While retaining the main rainpattern feature information, it reduces the static rainpattern detection network parameters and computational load, simplifies the operation, constructs the processing speed is fast, and improves the generalization ability of the depth network model in real-time.

The invention discloses a dual-channel single image fine rain removal method, and provides a new dual-channel residual dense block module (dual-channel hybrid block for short) for accurately obtaininga negative residual rain streak feature map, which comprises a residual path and a dense path. The residual path is used for reusing a common rain streak feature map from a front layer of the deep convolutional network, and the dense path can be used for exploring a new rain streak feature map. On the basis of a dual-channel mixing block, a cascaded dual-channel mixing block based on the dual-channel mixing block is constructed for rain streak feature extraction. In order to connect the characteristics of different scales, the method also adopts the idea of multi-stream branches, and data link channels are arranged between the multi-stream branches and between streams based on dual-channel residual dense blocks for rain streak characteristic information sharing. After a rain streak feature map of multiple sensing domains is obtained and a rough negative residual rain streak feature map is obtained through a convolutional neural network, a final accurate rain-removed image can be finally obtained through fine tuning.