CN110782414A - A dark-light image denoising method based on densely connected convolution - Google Patents

Abstract

The invention relates to a dim light image denoising method based on dense connection convolution, which comprises the following steps: constructing a dense connection denoising convolutional neural network model and training the dense connection denoising convolutional neural network model; acquiring an image to be processed, and preprocessing the image to be processed to obtain a target image to be processed; and processing the target image to be processed through the trained dense connection denoising convolutional neural network model to obtain a denoised image. The invention can keep a large amount of image detail information even when processing the image in a dim light or backlight scene, obtains a high-quality de-noising image and achieves a good de-noising effect.

Description

A dark-light image denoising method based on densely connected convolution

technical field

The invention relates to the technical field of digital image processing, in particular to a dark-light image denoising method based on densely connected convolution.

Background technique

Image noise refers to unnecessary or redundant interference information existing in image data. The existence of noise seriously affects the quality of the image, and noise reduction is particularly important.

The existing image denoising methods generally use the local information of the image for smoothing, or divide the image into blocks of a certain size, and combine two-dimensional image blocks with similar structures to form a three-dimensional image block according to the similarity between the image blocks. Then use the method of joint filtering to process these three-dimensional arrays, and return the processed results to the original image through inverse transformation, thereby obtaining the denoised image.

However, when these two image denoising methods deal with images in dark light or backlight scenes, the images often lose a lot of detail information, cannot obtain high-quality denoised images, and cannot achieve good denoising effects.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to propose a dark-light image denoising method based on densely connected convolution, which realizes that even when processing images in dark-light or backlight scenes, a large amount of image detail information can be retained, and high-quality images can be obtained. High-quality denoised images to achieve good denoising effects.

The present invention adopts the following scheme to realize: a dark light image denoising method based on dense connection convolution, comprising the following steps:

Build a densely connected denoising convolutional neural network model and train it;

Obtaining an image to be processed, and preprocessing the image to be processed to obtain a target image to be processed;

The target image to be processed is processed through the trained densely connected denoising convolutional neural network model to obtain a denoised image.

Further, the construction of the densely connected denoising convolutional neural network model is specifically: the densely connected denoising convolutional neural network model includes eight densely connected convolutional blocks, the four-channel image is used as the input of the network, and multiple The processing of densely connected convolution blocks; in which, in the first four blocks, each block is subjected to downsampling, and in the last four blocks, it is connected with the previous results of the same size to perform an upsampling, and then enters the block processing ; Finally output a 3-channel RGB image.

Further, each densely connected convolutional block includes four convolutional layers, each convolutional layer is connected to each subsequent convolutional layer by skipping within the block, and the last convolutional layer is a 1*1 convolutional layer, The output of this 1*1 convolutional layer is added to the input of the block to obtain the output of the current densely connected convolutional block.

Preferably, the combination of blocks of different sizes is to downsample the size of the input block by 1/2 times each time, and finally upsample by 2 times each time until it returns to the input size. In the same size block, the previous The output of the block is connected to the following block.

Further, the training of the densely connected denoising convolutional neural network model is specifically:

Construct an experimental training data set, which includes multiple sets of exposure images, each set of exposure images includes short-exposure images and long-exposure images of the same scene; use the constructed densely connected convolutional neural network model to train the experimental training data set , where the loss function is the average value of the absolute value of the difference between the network output and the long exposure image, and the training target is the minimum loss function.

Further, the preprocessing of the to-be-processed image is specifically:

processing the to-be-processed image into a plurality of single-color channel images;

The multiple single-color channel images are spliced to obtain the target image to be processed.

Further, the denoising convolutional neural network model of the trained dense connection is used to process the target image to be processed, and the obtained denoising image is specifically: the neural network model is convolved through some dense connections, and the image is subjected to different scales. The above processing outputs the image after denoising.

Compared with the prior art, the present invention has the following beneficial effects: the present invention adopts the densely connected convolutional neural network constructed to process the input noise image, and realizes that even when processing the image in the dark light or backlight scene, the Retain a lot of image detail information, get high-quality denoised images, and achieve good denoising effects.

Description of drawings

FIG. 1 is a design diagram of a neural network block according to an embodiment of the present invention.

FIG. 2 is an overall design diagram of a densely connected neural network model according to an embodiment of the present invention.

FIG. 3 is an original image of an embodiment of the present invention.

FIG. 4 is an example of restoring the image of FIG. 3 by using an existing method according to an embodiment of the present invention.

FIG. 5 is the image of FIG. 3 restored by the method of the embodiment of the present invention.

FIG. 6 is an original enlarged image in the box in FIG. 3 according to an embodiment of the present invention.

FIG. 7 is an enlarged image within the box in FIG. 3 that is restored by using an existing method according to an embodiment of the present invention.

FIG. 8 is an enlarged image within the box in FIG. 3 according to an embodiment of the present invention, which is restored by the method of this embodiment.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

This embodiment provides a dark-light image denoising method based on densely connected convolution, including the following steps:

Build a densely connected denoising convolutional neural network model and train it;

Obtaining an image to be processed, and preprocessing the image to be processed to obtain a target image to be processed;

The target image to be processed is processed through the trained densely connected denoising convolutional neural network model to obtain a denoised image.

In this embodiment, as shown in FIG. 2 , the construction of the densely connected denoising convolutional neural network model is specifically: the densely connected denoising convolutional neural network model includes eight densely connected convolutional blocks, and the four-channel The image is used as the input of the network to process multiple densely connected convolution blocks; among them, in the first four blocks, for each block, downsampling is performed again, and in the last four blocks, it is first connected with the previous results of the same size Perform an upsampling, and then enter the block processing; finally output a 3-channel RGB image.

In this embodiment, as shown in Figure 1, each densely connected convolutional block includes four convolutional layers, each convolutional layer is connected to each subsequent convolutional layer by skipping within the block, and the last convolutional layer The layer is a 1*1 convolutional layer, and the output of the 1*1 convolutional layer is added to the input of the block to obtain the output of the current densely connected convolutional block.

Preferably, the combination of blocks of different sizes is to downsample the size of the input block by 1/2 times each time, and finally upsample by 2 times each time until it returns to the input size. In the same size block, the previous The output of the block is connected to the following block.

In this embodiment, the training of the densely connected denoising convolutional neural network model is specifically:

Construct the experimental training data set, which includes multiple sets of exposure images, each set of exposure images includes short-exposure images and long-exposure images of the same scene; that is, in the image sample set uploaded by the user, each set of exposure images includes short-exposure images of the same scene. exposure images as well as long exposure images. A short exposure image of the same scene refers to an image taken under the same conditions as the shooting scene, light conditions (such as backlight, front light, dark light, etc.), exposure time, sensitivity, exposure, etc., and a long exposure image of the same scene. Refers to the image taken under the same scene as the short-exposure image, and other conditions are significantly better than the short-exposure image; the constructed densely connected convolutional neural network model is used to train the experimental training data set, in which the loss The function is the average value of the absolute value of the difference between the network output and the long exposure image, and the training goal is to minimize the loss function.

In this embodiment, the preprocessing of the to-be-processed image is specifically:

processing the to-be-processed image into a plurality of single-color channel images;

The multiple single-color channel images are spliced to obtain the target image to be processed.

Particularly, the terminal can convert the acquired channel mode of the image to be processed into multiple single-color channel images by calling a preset function. The channel that saves the color information of the image is called the color channel, and each color channel stores the information of the color elements in the image. A color channel image composed of information of only one color element is a single color channel image. For example, in an RGB mode image, R is a red channel, G is a green channel, and B is a blue channel.

For example, the image to be processed obtained by the terminal is a raw image (RAW Image Format, RAW), that is, the raw image information that is unprocessed and uncompressed. At this time, the terminal calls the preset function to convert the multi-color single-channel mode of the raw image. for multiple single-color channel images. Specifically, the terminal extracts each color in the original image by calling the preset function, and generates multiple single-color channel images. Wherein, the preset function may adopt a function in the prior art.

In this embodiment, the denoising convolutional neural network model that has been trained to process the target image to be processed to obtain a denoised image is specifically: the neural network model convolutions the image through some dense connections Perform processing on different scales, and output the image after noise removal.

By using the method of this embodiment, even when processing an image in a dark light or backlight scene, a large amount of image detail information can be retained, a high-quality denoised image can be obtained, and a good denoising effect can be achieved.

In particular, in order to verify the effectiveness of this embodiment, the denoising processing of the existing method and the method of this embodiment is used for the original image shown in FIG. 3 , and the results are shown in FIGS. 4 to 8 . FIG. 4 is an image denoised by using the existing method, and FIG. 5 is an image denoised by the method of this embodiment. In order to display more clearly, in this embodiment, the image in the box in FIG. 3 is enlarged, as shown in FIG. 6 , FIG. 7 is the image in the box after denoising using the prior art, and FIG. The image inside the denoised box. It can be seen from the drawings that the denoising effect of the method in this embodiment is obviously better than that of the prior art, and the obtained image denoising effect is better.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowcharts and/or block diagrams, and combinations of flows and/or blocks in the flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions An apparatus implements the functions specified in a flow or flows of the flowcharts and/or a block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

The above are only preferred embodiments of the present invention, and are not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications to equivalent changes. Example. However, any simple modifications, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solutions of the present invention still belong to the protection scope of the technical solutions of the present invention.

Claims (6)

1. a dark light image denoising method based on densely connected convolution, is characterized in that, comprises the following steps: Build a densely connected denoising convolutional neural network model and train it; Obtaining an image to be processed, and preprocessing the image to be processed to obtain a target image to be processed; The target image to be processed is processed through the trained densely connected denoising convolutional neural network model to obtain a denoised image. 2 . The method for denoising a dark light image based on densely connected convolution according to claim 1 , wherein the constructing a densely connected denoising convolutional neural network model is specifically: the densely connected denoising volume. 3 . The product neural network model includes eight densely connected convolutional blocks, and the four-channel image is used as the input of the network to process multiple densely connected convolutional blocks; among them, in the first four blocks, for each block, the next step is performed again. Sampling, in the last four blocks, first connect with the previous results of the same size to perform an upsampling, and then enter the block processing; finally output a 3-channel RGB image. 3. A method for denoising dark light images based on densely connected convolution according to claim 2, wherein each densely connected convolutional block comprises four convolutional layers, and each convolutional layer passes through the block in the block. The inner skip is connected to each subsequent convolutional layer. The last convolutional layer is a 1*1 convolutional layer. The output of the 1*1 convolutional layer is added to the input of the block to obtain the output of the current densely connected convolutional block. 4. a kind of dark light image denoising method based on dense connection convolution according to claim 1, is characterized in that, the training of described dense connection denoising convolutional neural network model is specifically: Construct an experimental training data set, which includes multiple sets of exposure images, each set of exposure images includes short-exposure images and long-exposure images of the same scene; use the constructed densely connected convolutional neural network model to train the experimental training data set , where the loss function is the average value of the absolute value of the difference between the network output and the long exposure image, and the training target is the minimum loss function. 5. The method for denoising a dark light image based on densely connected convolution according to claim 1, wherein the preprocessing of the to-be-processed image is specifically: processing the to-be-processed image into a plurality of single-color channel images; The multiple single-color channel images are spliced to obtain the target image to be processed. 6. a kind of dark light image denoising method based on dense connection convolution according to claim 1, it is characterized in that, described through training dense connection denoising convolutional neural network model to process the target image to be processed , to obtain a denoised image specifically: the neural network model processes the image at different scales through some densely connected convolutions, and outputs a denoised image.

Images