CN104537622A - Method and system for removing raindrop influence in single image - Google Patents

Abstract

The present invention provides a method and system for removing the influence of raindrops in a single image. The method includes: decomposing the image to be processed based on the empirical mode decomposition method, extracting the high-frequency part of the image to be processed, and forming a High-frequency feature map of high-frequency part information; Identify the edge of the image element in the image to be processed to obtain a feature contour map; Use image morphology operations to process the image area within the edge to obtain a first intermediate image; From the image Subtracting the first intermediate image from the high-frequency feature map to obtain a rain line feature map; subtracting the image to be processed from the rain line feature map to obtain a rain-removed image. The invention effectively improves the image processing speed.

Description

Method and system for removing the influence of raindrops in a single image

technical field

The invention relates to digital image processing technology, in particular to a method and system for removing the influence of raindrops in a single image.

Background technique

Due to its many advantages including automation, intelligence, and high efficiency, outdoor computer vision systems are widely used in military defense, medical technology, intelligent transportation and other fields. But severe weather can seriously affect its performance and even cause it to fail completely. Therefore, an effective method to eliminate the influence of severe weather is essential for an all-weather outdoor vision system. In many severe weather conditions, due to the large particle (raindrop) radius and other complex physical characteristics of rain, it will have a greater impact on the quality of the image captured by the vision system. Image raindrop removal technology uses the physical and frequency characteristics of rain to identify and remove raindrops in images. It can not only significantly improve the image quality, but also facilitate the further processing of the image. Therefore, image raindrop removal technology has become an indispensable key technology in the field of computer vision.

In recent years, research on raindrop detection and removal in images has become a hot topic. In 2003, Starik et al. first proposed the raindrop removal strategy of time-domain average value. The author believes that in the video image sequence, the influence of raindrops on pixels only exists in a few frames, so it can be removed by directly averaging the video frames. The original image of the rain effect. Unfortunately, they did not experimentally validate the method. Garg and Nayar were the first to use the dynamic and photometric properties of rain (K.Garg and S.K.Nayar, "Detection and removal of rain from videos," in Proc.IEEE Conf.Comput.Vis.Pattern Recognit., Jun.2004, vol.1 , pp.528–535), established two models respectively, and proposed methods for detecting and removing rain based on these two models. For the dynamic model of rain, it shows that the rain has time-domain correlation in its falling direction; for the photometric model, it is divided into static rain and dynamic rain models. For static raindrops, its brightness is significantly higher than the background it covers; for dynamic raindrops (rainlines), its brightness is determined by static raindrop brightness, background brightness and camera exposure time. Afterwards, the author proposed a method of using the frame difference method for initial detection of raindrops, using two characteristics for false detection removal, and finally using the front and rear frame image information for raindrop removal. While this method performs well, it cannot handle heavily out-of-focus (distant) rain, rain on bright backgrounds, and rain changes. In 2006 Zhang et al. (Zhang X P, Li H, Qi Y Y, Leow W K, Ng T K. Rain removal in video by combining temporal and chromatic properties. In: Proceedings of the 2006 International Conference on Multimedia and Expo. Toronto, Canada: IEEE , 2006.461: 464) used the temporal distribution and color characteristics of rain. Since the histogram of the time domain distribution of rain shows two peaks (representing the brightness of raindrops and the background brightness respectively) and approximately constitutes a Gaussian mixture model, the unsupervised learning method - K-means clustering can effectively separate them. Afterwards, the author found that the inter-frame RGB value changes of pixels affected by raindrops are basically the same, so false detections can be further removed. The experimental effect of this method is good, but the clustering method is used to distinguish the raindrops and the background in the whole video, the calculation efficiency is not high, and it cannot be processed in real time. In 2007, Barnum et al. (Barnum P C, Narasimhan S G, Kanade T.Analysis of rain and snow in frequency space.Internatio-nal Journal of ComputerVision, 2010,86(2:3):256:274) noticed that most of the previous methods Heavy reliance is placed on the extraction of clear rainlines, which make it reasonable to analyze rain in the frequency domain because rainlines cause repetitive patterns. The author establishes a Gaussian model to approximate the influence of rain, and detects raindrops by finding the proportion of the model in the three-dimensional Fourier transform, and then iterates to remove the rain, and finally inversely transforms it to the video image. Experimental results show that this method has better processing performance, but the time complexity of this method is too high, and for the processing of inconspicuous rain and rain changes, it will experience significant performance degradation.

The above rain removal methods based on a single image can only process grayscale images, and the method takes a long time. For example, the latest optimization algorithm (the method of Chen et al.) takes more than 100s to process a specific single image. At the same time, the output image will appear blurred to a certain extent.

Based on the time complexity of the single image raindrop removal technology in the prior art is too high, which is not conducive to the promotion of the method, it is necessary to further improve the raindrop removal technology in the image.

Contents of the invention

Based on this, it is necessary to address the problems existing in the prior art and provide a method and system for removing the influence of raindrops in a single image, which can process color images and effectively improve the image processing speed.

A method for removing the influence of raindrops in a single image, comprising:

Decomposing the image to be processed based on the empirical mode decomposition method, extracting the high-frequency part of the image to be processed, and forming a high-frequency feature map reflecting the information of the high-frequency part;

Identifying edges of image elements in the image to be processed to obtain a feature contour map;

processing the image region within the edge by using image morphology operations to obtain a first intermediate image;

subtracting the first intermediate image from the high-frequency feature map to obtain a rainline feature map;

The image to be processed is subtracted from the rain line feature map to obtain a rain-removed image.

In one embodiment, the method also includes:

The feature contour map is superimposed on the derained image to obtain a repaired image.

In one of the embodiments, the process of decomposing the image to be processed and extracting the high-frequency part of the image to be processed based on the empirical mode decomposition method includes the following steps:

Mapping the image to be processed to the XOY plane, the gray value of the pixel corresponding to the image to be processed as the Z coordinate;

Identifying local maxima and minima of the image to be processed by a morphological method to obtain a plurality of scattered maxima and minima;

Carrying out triangulation of the planar point set to the plurality of scattered maximum value points and minimum value points respectively, and then interpolating and smoothing to obtain the maximum value envelope surface and the minimum value envelope surface;

calculating the mean value of the maximum value envelope surface and the minimum value envelope surface;

Subtracting the mean value from the gray value of each pixel in the image to be processed to obtain a decomposed image;

Judging whether the decomposed image satisfies the screening end condition, if so, then output the decomposed image as the high-frequency part; if not, return to the process of identifying the local maximum of the image to be processed by the morphological method value and minimum value steps.

In one of the embodiments, the step of outputting the decomposed image as the high-frequency part further includes:

subtracting the decomposed image from the image to be processed to obtain the currently processed image;

Judging whether the currently processed image satisfies the image decomposition end condition, and if so, outputting the decomposed image as the high-frequency part;

If not, perform the next decomposition, return to the step of identifying the local maximum and minimum values of the image to be processed by the morphological method, until the end condition of the image decomposition is satisfied, and output all the results obtained by multiple decompositions The decomposed image is output as the high frequency part.

In one of the embodiments, in the process of identifying the edges of the image elements in the image to be processed to obtain the feature contour map, edge detection is performed on the image based on the image grayscale using a Prewitt operator.

In one of the embodiments, the process of processing the image region within the edge by using image morphology operation includes: filling the connected image region of the edge based on an image processing erosion operation.

In one of the embodiments, the process of subtracting the first intermediate image from the high-frequency feature map to obtain the rainline feature map includes:

Inverting the first intermediate image to obtain a second intermediate image;

The second intermediate image is multiplied by the high-frequency feature map, and the intersection of the second intermediate image and the high-frequency feature map is extracted to form the rain line feature map.

In one of the embodiments, the process of subtracting the image to be processed from the rain line feature map to obtain the image after rain removal includes:

Inverting the rainline feature map to obtain a third intermediate image;

Multiplying the image to be processed by the third intermediate image, extracting the intersection of the third intermediate image and the image to be processed, to form the image after rain removal.

A system for removing the influence of raindrops in a single image, comprising:

The image decomposition module is used to decompose the image to be processed based on the empirical mode decomposition method, extract the high-frequency part of the image to be processed, and form a high-frequency feature map reflecting the information of the high-frequency part;

An edge detection module, configured to identify the edge of the image element in the image to be processed, and obtain a feature contour map;

A filling module, configured to process the image region within the edge by using image morphology operations to obtain a first intermediate image;

A first computing module, configured to subtract the first intermediate image from the high-frequency feature map to obtain a rainline feature map; and

The second computing module is used to subtract the image to be processed from the rain line feature map to obtain the image after rain removal.

In one of the embodiments, the system also includes:

The superposition processing module is used to superimpose the feature contour map on the derained image to obtain a repaired image.

Based on the above method and system, the present invention obtains the high-frequency part of the image through the image decomposition technology based on empirical mode decomposition, and then uses the image edge recognition algorithm to subtract the results of the two, so as to finally obtain the pixels affected by the rain. The brightness of affected pixels is relatively high, and finally the rain-removed image is obtained by subtracting it from the original image. The invention can effectively improve the visual effect of the image affected by rain, can process color images, and improves the computing speed. Using the method of the invention can reduce the single image processing time by about 50%.

Description of drawings

Fig. 1 is a schematic flow sheet of an embodiment of the inventive method;

Fig. 2 is a schematic flow sheet of another embodiment of the inventive method;

Fig. 3 is a schematic structural diagram of an embodiment of the system of the present invention;

Fig. 4 is the rendering of image to be processed;

Fig. 5 is an effect diagram of a repaired image in an embodiment of the present invention.

Detailed ways

Based on the image deraining technology in the field of machine vision, the present invention proposes a new image deraining method, which obtains the high frequency part of the image through the image decomposition technology based on empirical mode decomposition, and then uses the image edge recognition algorithm, the two The results are subtracted, so as to finally obtain the pixels affected by the rain. Because the brightness of the pixels affected by the rain is relatively high, finally by subtracting it from the original image, the rain-removed image is obtained. The present invention can effectively improve the visual effect of the image affected by the rain. , simplify the calculation process of image processing, and speed up the efficiency of image processing. Various embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in FIG. 1 , the present invention provides a method for removing the influence of raindrops in a single image, which includes the following steps.

In step 100, the image to be processed is decomposed based on the empirical mode decomposition method, the high frequency part of the image to be processed is extracted, and a high frequency feature map reflecting the information of the high frequency part is formed. The empirical mode decomposition method mentioned here refers to a new non-stationary signal analysis method, which has the advantages of locality and adaptability. In one embodiment of the present invention, a two-dimensional Empirical Mode Decomposition (BEMD, Bidimensional Empirical Mode Decomposition) method is used to decompose the image to be processed. As another example, in a preferred embodiment of the present invention, as shown in FIG. 2, the execution process of the above step 100 includes the following steps:

Step 101, mapping the image to be processed to the XOY plane, and using the gray value of the pixel corresponding to the image to be processed as the Z coordinate;

Step 102, identifying local maxima and minima of the image to be processed by a morphological method, and obtaining a plurality of scattered maxima and minima;

Step 103, performing triangulation of the plane point set on the above-mentioned scattered maximum and minimum points, and then interpolating and smoothing to obtain the maximum envelope surface and the minimum envelope surface;

Step 104, calculating the mean value of the above-mentioned maximum value envelope surface and minimum value envelope surface;

Step 105, subtracting the mean value from the gray value of each pixel in the image to be processed to obtain a decomposed image;

Step 106, judging whether the above-mentioned decomposed image satisfies the screening end condition, and the screening end condition is specifically the inspection of the zero-crossing condition and the mean value condition. If the number of extremum points (the extremum points include maxima and minima) is equal to or differs from the number of crossing zero points by one and the above-mentioned mean formed by local maxima is zero, then step 109 is executed, if not , return to the step 102 of identifying the local maximum and minimum values of the image to be processed through the morphological method.

Step 109, use the above-mentioned decomposed image as the image detail information obtained in this decomposition process, if the above-mentioned decomposed image satisfies the screening end condition, it can be used as the output of the above-mentioned high-frequency part, if there are multiple image detail information obtained by multiple decomposition processes, Then, the decomposition images obtained by the above multiple decompositions are superimposed as the above high-frequency feature map. As another example, in another embodiment of the present invention, before the above-mentioned step 109 of outputting the above-mentioned decomposed image as the above-mentioned high-frequency part, it further includes:

Step 107, subtracting the above decomposed image from the above image to be processed to obtain the currently processed image;

Step 108, judge whether the above-mentioned currently processed image satisfies the image decomposition end condition, the image decomposition end condition is specifically whether each layer of image detail information has no more than one extreme point, and if so, perform the above-mentioned step 109: take the above-mentioned decomposed image as The above-mentioned high-frequency part output;

If not, add one to the number of iterations for the next decomposition, and return to the step 102 to identify the local maxima and minima of the image to be processed through the morphological method until the above image decomposition end condition is satisfied , output the above-mentioned decomposed images respectively obtained by multiple decompositions, and output them as the above-mentioned high-frequency part.

Furthermore, in one embodiment of the present invention, the step of forming a high-frequency feature map embodying high-frequency part information is as follows: if the above-mentioned high-frequency part only includes a decomposed image obtained by one decomposition process, the decomposed image is used as The above-mentioned high-frequency feature map; if the above-mentioned high-frequency part includes decomposition images respectively obtained by multiple decomposition processes, the above-mentioned decomposition images obtained by multiple decomposition processes are superimposed to form the above-mentioned high-frequency feature map.

In the above embodiment, more detailed information of the image can be obtained after multiple iterative decompositions using the above steps, and high frequency parts of the image, including raindrops and object boundaries, can be obtained.

In step 200, edges of image elements in the image to be processed are identified to obtain a feature contour map. The feature contour map here is preferably a binary image obtained by processing a grayscale image. In one embodiment of the present invention, in this step, in the process of identifying the edge of the image element in the image to be processed to obtain the feature contour map, the edge detection algorithm based on the gradient is used to detect the edge of the image, preferably, based on the image grayscale using The Prewitt operator detects the edge of the image. The Prewitt operator is more suitable for the situation where the gray value of the edge of the image is sharp and the image noise is relatively small, and its processing speed is faster. Of course, the present invention is not limited to only adopting this method for edge detection, for example, one of methods such as Roberts edge operator, Sobel operator, Laplacian operator, and Canny operator can also be used for edge detection of the image to be processed.

In step 300, the image region inside the above-mentioned edge is processed by image morphology operation to obtain a first intermediate image. In a preferred embodiment of the present invention, the process of processing the image area within the edge by using image morphology operation includes: filling the connected image area of the edge based on image processing erosion operation. Morphological processing in digital image processing refers to the use of digital morphology as a tool to extract image components that are useful for expressing and describing the shape of a region, such as boundaries, skeletons, and convex hulls, as well as for preprocessing or postprocessing. Morphological filtering, thinning and pruning, etc. Among them, for the erosion operation, it can add pixels to the object boundary in the image, and even perform connected region filling. In this embodiment, based on a connected region filling algorithm of an image processing erosion operation, hole filling is performed on the feature contour map to obtain the above-mentioned first intermediate image.

In step 400, the above-mentioned first intermediate image is subtracted from the above-mentioned high-frequency feature map to obtain a rain line feature map. In this step, the image arithmetic operation method is used to process the two images, which improves the operation speed. Preferably, in one embodiment of the present invention, the above-mentioned process of subtracting the above-mentioned first intermediate image from the above-mentioned high-frequency feature map includes the following steps:

First, invert the above-mentioned first intermediate image to obtain a second intermediate image;

Secondly, the above-mentioned second intermediate image is multiplied by the above-mentioned high-frequency feature map, and the intersection of the above-mentioned second intermediate image and the above-mentioned high-frequency feature map is extracted to form the above-mentioned rain line feature map.

In step 500, the above image to be processed is subtracted from the above rain line feature map to obtain an image after rain removal. In this step, the image arithmetic operation method is used to process the two images, which improves the operation speed. Preferably, in one embodiment of the present invention, the above-mentioned process of subtracting the above-mentioned image to be processed from the above-mentioned rain line feature map to obtain the image after rain removal includes the following steps:

First, invert the above rainline feature map to obtain the third intermediate image;

Secondly, multiply the third intermediate image by the image to be processed, extract the intersection of the third intermediate image and the image to be processed, and form the image after rain removal.

Based on the above-mentioned embodiment, in one embodiment of the present invention, as shown in Figure 1, the above-mentioned method also includes:

Step 600, superimposing the above-mentioned feature contour map on the above-mentioned image after rain removal, to obtain a repaired image.

Based on the deraining result map obtained in the above step 500, since some required image features are included in the rain line feature map, the deraining result map can be added to the feature contour map identified by the corresponding contour recognition operator , that is, the repaired rain-removed image can be obtained, the accuracy of image processing results can be improved, and the visual effect of the image can be improved.

Fig. 1 is a schematic flow chart of a method according to an embodiment of the present invention. It should be understood that although the various steps in the flow chart of FIG. 1 are displayed sequentially as indicated by the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in Figure 1 may include multiple sub-steps or multiple stages, these sub-steps or stages are not necessarily executed at the same time, but may be executed at different times, and the execution order is not necessarily performed sequentially, but may be combined with other steps or sub-steps or stages of other steps or the order of execution may be exchanged. In the specific description above, each of the above embodiments only elaborates on the implementation of the corresponding steps, and if the logic is not contradictory, the above-mentioned embodiments can be combined with each other to form a new technical solution, and the new The technical solution is still within the disclosure scope of this specific embodiment.

As shown in Figure 3, based on the above method, the present invention also provides a system 800 for removing the influence of raindrops in a single image, which includes:

An image decomposition module 801, configured to decompose the image to be processed based on the empirical mode decomposition method, extract the high-frequency part of the image to be processed, and form a high-frequency feature map reflecting the information of the high-frequency part;

An edge detection module 802, configured to identify edges of image elements in the image to be processed, and obtain a feature contour map;

A filling module 803, configured to use image morphology operations to process the image area within the edge to obtain a first intermediate image;

A first computing module 804, configured to subtract the first intermediate image from the high-frequency feature map to obtain a rainline feature map; and

The second computing module 805 is configured to subtract the image to be processed from the rain line feature map to obtain an image after rain removal.

Based on the above-mentioned embodiment, as shown in FIG. 3, in one embodiment of the present invention, the above-mentioned system further includes the following functional modules:

The superposition processing module 806 is configured to superimpose the feature contour map on the derained image to obtain a repaired image.

The above-mentioned functional modules 801 to 806 are respectively used to execute the above-mentioned steps 100 to 600. For the specific implementation manner, please refer to the above-mentioned related descriptions about the steps 100 to 600, which will not be repeated here.

In one embodiment of the present invention, the above image decomposition module 801 includes:

A mapping unit, configured to map the image to be processed to an XOY plane, where the gray value of the pixel corresponding to the image to be processed is used as the Z coordinate;

An identification unit, configured to identify the local maximum and minimum of the image to be processed by a morphological method, and obtain a plurality of scattered maximum and minimum points;

An interpolation unit, used to perform triangulation of the plane point set on the plurality of scattered maximum points and minimum points, and then interpolate and smooth to obtain a maximum envelope surface and a minimum envelope surface;

a mean calculation unit, configured to calculate the mean of the maximum envelope surface and the minimum envelope surface;

a decomposition unit, configured to subtract the mean value from the gray value of each pixel in the image to be processed to obtain a decomposed image;

The first judging unit is used to judge whether the decomposed image satisfies the screening end condition, and if so, output the decomposed image as the high-frequency part; if not, return to call the recognition unit.

In one embodiment of the present invention, the above-mentioned image decomposition module 801 also includes:

a second decomposing unit, configured to subtract the decomposed image from the image to be processed to obtain the currently processed image;

The second judging unit is used to judge whether the currently processed image satisfies the end condition of image decomposition, if so, output the decomposed image as the high-frequency part; if not, perform the next decomposition, and return to call the above identification The unit, until the end condition of the image decomposition is satisfied, outputs the decomposed image obtained by multiple decompositions as the high-frequency part output. In another embodiment of the present invention, the image decomposition module 801 further includes: a high-frequency feature map forming unit, configured to use the decomposed image as the above-mentioned high-frequency feature map if the above-mentioned high-frequency part only includes a decomposed image obtained by one decomposition process If the above-mentioned high-frequency part includes the decomposition images respectively obtained by multiple decomposition processes, then superimpose the above-mentioned decomposition images obtained by multiple decompositions to form the above-mentioned high-frequency feature map.

Each functional unit in the above-mentioned image decomposition module 801 is respectively used to execute step 101 to step 109 in FIG. 2 , so its specific implementation method can refer to the above-mentioned relevant description on steps 101 to 109 , and will not be repeated here.

In one embodiment of the present invention, the first computing module 804 includes:

a first inversion unit, configured to invert the first intermediate image to obtain a second intermediate image;

A first multiplying unit, configured to multiply the second intermediate image by the high-frequency feature map, extract an intersection of the second intermediate image and the high-frequency feature map, and form the rain line feature map.

In one embodiment of the present invention, the second computing module 805 includes:

The second inversion unit is used to invert the rain line feature map to obtain a third intermediate image;

The second multiplying unit is configured to multiply the third intermediate image by the image to be processed, extract the intersection of the third intermediate image and the image to be processed, and form the image after rain removal.

Each functional module in the system 800 for removing the influence of raindrops in the above-mentioned single image is used to perform each step of the method for removing the influence of raindrops in the above-mentioned single image shown in FIG. Explanations are not repeated here.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on such an understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is carried on a non-volatile computer-readable storage carrier (such as ROM, magnetic disk, optical disk, server storage space), including several instructions to make a terminal device (which can be a mobile phone, computer, server, or network device, etc.) execute the system structure and method described in various embodiments of the present invention .

In summary, the present invention proposes a new image deraining method in this paper, which obtains the high-frequency part of the image through the image decomposition technology based on empirical mode decomposition, and then uses the image edge recognition algorithm, and the results of the two are similar. Subtract, so that the pixels affected by the rain are finally obtained. Since the pixels affected by the rain have higher brightness, finally, the rain-removed image is obtained by subtracting it from the original image. The invention can effectively improve the visual effect of images affected by rain, can process color images, and improves the computing speed. Furthermore, the present invention uses empirical mode decomposition to decompose images, and performs intersection processing between images based on image arithmetic operations, which has better performance than other algorithms, and greatly reduces the time required for rain removal, which is far lower than processing time by traditional methods. The invention also overcomes the disadvantage that the single image rain removal algorithm based on sparse coding can only process grayscale images, can process color images, and can obtain better results. For example, for the image to be processed shown in Figure 4, after processing the above steps 100 to 600 based on the two-dimensional empirical mode decomposition method and the Prewitt edge recognition algorithm, the effect diagram shown in Figure 5 is obtained, and an image is processed The time is 25.473 seconds.

The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but should not be construed as limiting the patent scope of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (10)

1. remove a method for raindrop impact in single image, it comprises:

Based on empirical mode decomposition method, picture breakdown is carried out to pending image, extract the HFS of described pending image, form the high-frequency characteristic figure embodying described HFS information;

Identify the edge of pictorial element in described pending image, obtain feature contour figure;

Utilize morphological image to operate to process described intramarginal image-region, obtain the first intermediate image;

From described high-frequency characteristic figure, deduct described first intermediate image, obtain rain line features figure;

Described pending image and described rain line features figure are subtracted each other, obtains the image after removing rain.

2. remove the method for raindrop impact in single image according to claim 1, it is characterized in that, described method also comprises:

Described remove rain after image on superpose described feature contour figure, obtain repair after image.

3. the process of the HFS remove the method for raindrop impact in single image according to claim 1, it is characterized in that, describedly carry out picture breakdown based on empirical mode decomposition method to pending image, extracting described pending image comprises the following steps:

By described pending image mapped to XOY plane, the gray-scale value of described pending image respective pixel is as Z coordinate;

Identified local maximum and the minimal value of described pending image by morphological method, obtain multiple scattered maximum point and minimum point;

Described multiple scattered maximum point and minimum point are carried out respectively to the triangulation of plane point set, then interpolation smoothing obtains maximum value envelope surface and minimal value envelope surface;

Calculate the average of described maximum value envelope surface and minimal value envelope surface;

The gray-scale value of each pixel in described pending image is deducted described average, obtains exploded view picture;

Judge that described exploded view similarly is no satisfied screening termination condition, if so, then described exploded view picture is exported as described HFS; If not, then return described in execution and identify the local maximum of described pending image and minimizing step by morphological method.

4. remove the method for raindrop impact in single image according to claim 3, it is characterized in that, described described exploded view picture also to be comprised before the step that described HFS exports:

From described pending image, deduct described exploded view picture, obtain when the image after pre-treatment;

When whether the image after pre-treatment meets picture breakdown termination condition, if then exported as described HFS by described exploded view picture described in judging;

If not, then decompose next time, return to perform and identify the local maximum of described pending image and minimizing step by morphological method, until meet described picture breakdown termination condition, export and repeatedly decompose the described exploded view picture obtained respectively, export as described HFS.

5. in single image according to claim 1, remove the method for raindrop impact, it is characterized in that, in the described pending image of described identification, the edge of pictorial element obtains in the process of feature contour figure, adopts Privett operator to carry out rim detection to image based on gradation of image.

6. in single image according to claim 1, remove the method for raindrop impact, it is characterized in that, described utilize morphological image operate the process that described intramarginal image-region processes is comprised: fill based on the Contiguous graphics region of image procossing etching operation to described edge.

7. remove the method for raindrop impact in single image according to claim 1, it is characterized in that, describedly from described high-frequency characteristic figure, deduct the process that described first intermediate image obtains rain line features figure comprise:

To described first intermediate image negate, obtain the second intermediate image;

Be multiplied with described high-frequency characteristic figure with described second intermediate image, extract the common factor of described second intermediate image and described high-frequency characteristic figure, form described rain line features figure.

8. remove in single image according to claim 1 raindrop impact method, it is characterized in that, described described pending image and described rain line features figure are subtracted each other to obtain go the process of the image after rain to comprise:

To described rain line features figure negate, obtain the 3rd intermediate image;

Be multiplied with described pending image with described 3rd intermediate image, extract the common factor of described 3rd intermediate image and described pending image, described in formation, remove the image after rain.

9. remove a system for raindrop impact in single image, it is characterized in that, described system comprises:

Picture breakdown module, for carrying out picture breakdown based on empirical mode decomposition method to pending image, extracts the HFS of described pending image, forms the high-frequency characteristic figure embodying described HFS information;

Edge detection module, for identifying the edge of pictorial element in described pending image, obtains feature contour figure;

Packing module, operates for utilizing morphological image and processes described intramarginal image-region, obtain the first intermediate image;

First computing module, for deducting described first intermediate image from described high-frequency characteristic figure, obtains rain line features figure; And

Second computing module, for described pending image and described rain line features figure being subtracted each other, obtains the image after removing rain.

10. remove the system of raindrop impact in single image according to claim 9, it is characterized in that, described system also comprises:

Overlap-add procedure module, for described remove rain after image on superpose described feature contour figure, obtain repair after image.