CN109712099A

CN109712099A - Sonar image equalization method based on SLIC and adaptive filtering

Info

Publication number: CN109712099A
Application number: CN201811474850.XA
Authority: CN
Inventors: 宋艳; 李沂滨; 何波
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2018-12-04
Filing date: 2018-12-04
Publication date: 2019-05-03
Anticipated expiration: 2038-12-04
Also published as: CN109712099B

Abstract

The invention belongs to the technical field of sonar image enhancement, and discloses a sonar image equalization method based on SLIC and adaptive filtering. Regular image blocks; 3. Unique pixel values for each image block, sort the pixel values, and obtain the median value of the pixels; 4. Update the gray value of the image block, and traverse all image blocks according to step 3; 5. Repeat the The sonar image is divided into different numbers of image blocks, and steps 3 and 4 are repeated respectively; 6. Average the equalized images in step 5 obtained multiple times to obtain the final gray-scale equalized image. The technical solution fully considers the grayscale similarity of adjacent superpixel image blocks and the grayscale balance of each superpixel image block, and the grayscale equalization effect is good, and has a significant effect on improving the accuracy of sonar image classification.

Description

Method is equalized based on the sonar image of SLIC and adaptive-filtering

Technical field

The present invention relates to sonar images to enhance technical field, more particularly to the sonar image based on SLIC and adaptive-filtering Equalization method.

Background technique

Sonar image classification be sub-sea floor targets identification important means, can be used for seafloor topography detection, submarine survey, The fields such as submarine pipeline tracking and state-detection.Currently, the unbalanced problem of gray scale in sonar image has seriously affected sonar figure As the effect of classification, so, solution sonar image gray scale is unbalanced to have a direct impact raising sonar image classification accuracy rate.

It is influenced by sea bottom complex environment, acoustic noise and towed body motion state, it is abnormal often to there is gray scale in sonar image Become.Existing sonar image grayscale equalization method is based primarily upon towed body motion state, these states are passed by the posture in towed body Sensor and height sensor provide.The method of sonar image rectification is paper Data of the Capus et al. in 2008 at present Correction for visualisation and classification of sidescan SONAR imagery intermediary The method to continue.It is influenced by ocean wave motion and sea bottom complex terrain, the towed body motion state that sensor provides is constantly in variation State, meanwhile, the sensitivity and accuracy of sensor itself have error, by these sensor data correction sonar images, Correction course can be made to have biggish calculation amount and calculate error.In the segmentation and classification of sonar image, unbalanced gray scale is to influence One main problem of sonar image segmentation accuracy.At present about in the research that sonar image is divided and classifies, gray scale is uneven Weighing apparatus problem is not considered separately, and people focus primarily on emphasis in the method for classification or segmentation, although can also obtain in this way compared with Good classification or segmentation effect, but algorithm design is more complex, process is also extremely time-consuming.

Simple linear iteration clusters (SLIC): Simple Linear Iterative Clustering.

Summary of the invention

It is an object of the invention in view of the above technical problems, obtain the sonar image of not tonal distortion, one kind is provided Method is equalized based on the sonar image of SLIC and adaptive-filtering, realizes the grayscale equalization of a variety of sonar images, is seabed Target acquisition and classification provide the sonar image of correction, super-pixel cluster are carried out to sonar image, based on adaptive-filtering to poly- The sonar image block that class divides is filtered, and by the way that above steps may be repeated multiple times, realizes the grayscale equalization of sonar image.Specifically Steps are as follows:

Step 1, the speckle noise in sonar image is filtered out；

Step 2, sonar image is divided into irregular image block using simple linear iterative clustering methods；

Step 3, it carries out pixel value respectively to each image block uniquely to change, sorted pixels value, obtains pixel median；

To image block P₀And its all adjacent image block { P₁,P₂,K,P_k, uniquely change their pixel value, obtains pixel Race { A₀,A₁,A₂,...,A_k, A_i(i=0,1 ..., k) indicate image block P_iWhat the pixel value of (i=0,1 ..., k) was uniquely changed As a result, removal P_iIn repetition pixel value；To A_iPixel value in (i=0,1 ..., k) is arranged according to sequence from small to large Sequence；Take A_iIntermediate value a_i(i=0,1 ..., k).

Step 4, image block gray value is updated, traverses all image blocks according to step 3；

If method particularly includes: a_i(i=1,2 ..., k) is equal to 0 or 255, then image block P_iPicture in (i=1,2 ..., k) Plain value all resets to 0 or 255；If a_i(i=1,2 ..., k) be not equal to 0, also be not equal to 255, and | a_i-a₀| (i=1,2 ..., K) it is less than given threshold, then uses P_i'=a₀± rand (X) updates image block P_i(i=1,2 ..., gray value k), rand (X) it indicates to generate a number at random in [1, X] range, updated image block is P_i'。

Step 5, sonar image is repeatedly divided into the image block of different number, repeats step 3 and step 4 respectively；

Step 6, the equalization image of the step 5 repeatedly obtained is averaged, obtains final grayscale equalization image.

Further: X=5.

Further: image block number is 3000.

Further: the image block number in step 5 is 2500 and/or 2000 and/or 1500 and/or 1000.

The technical program is based on simple linear iterative clustering methods and sonar image is divided into irregular super-pixel image Block is filtered each super-pixel image block using adaptive filter method, has fully considered neighbouring super pixels image block Grey similarity and each super-pixel image block gray scale balance, pass through local equalization realize image entirety equilibrium Change, after realizing grayscale equalization, can be obtained by preferably using simple image partition method (such as k means clustering algorithm) Segmentation result；This method does not use the attitude data of sonar towed body mounted, avoids inherent data present in calculating process Error.The sonar image obtained using this method, can solve the unbalanced problem of gray scale in sonar image, grayscale equalization effect It is good, there is positive effect to sonar image classification accuracy rate is improved.

Detailed description of the invention

Fig. 1: sonar image；

Fig. 2: simple linear iteration clusters the cluster result to Fig. 1；

Fig. 3: Fig. 1 grayscale equalization result；

Segmentation result of Fig. 4: the k means clustering algorithm to Fig. 1；

Segmentation result of Fig. 5: the k means clustering algorithm to Fig. 3.

Wherein, the white grid lines in Fig. 2 is the boundary of each cluster block.

Specific embodiment

1-5 with reference to the accompanying drawings specifically describes embodiments of the present invention.

Step 1, median filtering is carried out to sonar image, the size of filter multiplies 3 for 3, and the purpose of this step is to filter out sound Speckle noise in image.Median filtering is the common method in image filtering.

Step 2, it (is proposed within Achanta 2012) using simple linear iterative clustering methods, sonar image is divided into not The image block of rule.The size of sonar image is 419 × 317, K in Fig. 1_l=3000, K_lIndicate the number for the image block that SLIC is divided Amount.Fig. 2 is SLIC cluster result, and the white grid lines in figure is the boundary of each cluster block.

Step 3- step 5, to each image block that SLIC is obtained, using adaptive-filtering, this step is pass of the invention Key step.Detailed process are as follows:

A. pixel value is carried out respectively to each image block and its adjacent image block uniquely to change.To image block P₀And its institute There is adjacent image block { P₁,P₂,K,P_k, uniquely change their pixel value, obtains pixel race { A₀,A₁,A₂,...,A_k, A_i(i=0, 1 ..., k) indicate image block P_iThe result that the pixel value of (i=0,1 ..., k) is uniquely changed.If P_iPixel value be 4,7,0,0, 0,1,2,3,3,4, }, the result A that pixel value is uniquely changed_iFor { 4,7,0,1,2,3 }.

B. to A_iPixel value in (i=0,1 ..., k) is ranked up according to sequence from small to large.If A_iFor 4,7,0, 1,2,3 }, the result of sequence is { 0,1,2,3,4,7 }.

C. after sorting, the intermediate value a of pixel value is found_i(i=0,1 ..., k).A_iIntermediate value a_iIt is 2.5.

The processing of a-c step can reduce the unbalanced influence to intermediate value of gray value.If image block is that grey value profile is equal Weighing apparatus, then the maximum value and minimum value and intermediate value difference of its gray value are little.And if image block grey value profile is unbalanced, The possible gap of maxima and minima is larger.When gray value number less than normal is more, intermediate value is biased to lesser gray value；Work as gray scale Be worth number bigger than normal it is more when, intermediate value can be biased to biggish gray value.Uniqueization processing in a step can reduce this deviation.

If d. a_i(i=1,2 ..., k) is equal to 0 or 255, then image block P_iPixel value in (i=1,2 ..., k) is all heavy It is set to 0 or 255.

If a_i(i=1,2 ..., k) be not equal to 0, also be not equal to 255, and | a_i-a₀| (i=1,2 ..., k) it is less than setting Threshold value illustrates image block P_iWith image block P₀Gray value it is similar, according to formula (1) update image block P_i(i=1,2 ..., k) Gray value, rand (5) expression generate a number at random in [1,5] range.Updated image block is P_i'.After update, figure As block P_i' gray value and P₀Approximation realizes P₀Neighborhood gray value equalization.

P_i'=a₀±rand(5) (1)

E. all image blocks are traversed according to step a-d.

Step 6, using smaller K_lValue (such as K_l=2500, K_l=2000, K_l=1500, K_l=1000) step, is repeated 3- step 5, to different K_lThe equalization image that value obtains is averaged, and obtains final grayscale equalization image.The gray scale of Fig. 1 is equal Weighing apparatusization result is as shown in Figure 3.

By grayscale equalization, the contrast of target (the white high bright part of Fig. 3) is become apparent from sonar image, in Fig. 1 The unbalanced problem of gray scale effectively solved；After grayscale equalization, sonar image can be achieved with using simple dividing method Effectively segmentation.It is (white that sonar image (Fig. 1 and Fig. 3) using simple k means clustering algorithm is divided into target area by Fig. 4 and Fig. 5 Color), background area (grey) and shadow region (black).As can be seen from the results, by grayscale equalization, segmentation effect is more preferable.

Embodiment only illustrates technical solution of the present invention, rather than carries out any restrictions to it；Although with reference to the foregoing embodiments Invention is explained in detail, for those of ordinary skill in the art, still can be to previous embodiment institute The technical solution of record is modified or equivalent replacement of some of the technical features；And these modifications or substitutions, and The essence of corresponding technical solution is not set to be detached from the spirit and scope of claimed technical solution of the invention.

Claims

1. the sonar image equalization method based on SLIC and adaptive filtering, is characterized in that, comprises the following steps:

Step 1, filter out speckle noise in the sonar image;

Step 2, using a simple linear iterative clustering method to divide the sonar image into irregular image blocks;

Step 3: Unique pixel values are performed on each image block, pixel values are sorted, and median pixel values are obtained;

Step 4, update the gray value of the image block, and traverse all the image blocks according to step 3;

Step 5: Divide the sonar image into image blocks of different numbers for many times, and repeat step 3 and step 4 respectively;

In step 6, the equalized images obtained in step 5 are averaged to obtain a final gray-scale equalized image.

2. the sonar image equalization method based on SLIC and adaptive filtering according to claim 1, is characterized in that, the concrete method of step 3 is: to image block P ₀ and all its adjacent image blocks {P ₁ , P ₂ , K, P _k }, uniquely their pixel values, get the pixel family {A ₀ , A ₁ , A ₂ ,...,A _k }, A _i (i=0,1,...,k ) represents the unique result of the pixel value of the image block P _i (i=0,1,...,k), and removes the repeated pixel values in P _i ; for A _i (i=0,1,..., The pixel values in k) are sorted in ascending order; take the median value of A _i ( _i =0,1,...,k).

3. the sonar image equalization method based on SLIC and adaptive filtering according to claim 2, is characterized in that, the concrete method of step 4 is: if a _i (i=1,2,...,k) is equal to 0 or 255, the pixel values in the image block P _i (i=1,2,...,k) are reset to 0 or 255; if a _i (i=1,2,...,k) is not equal to 0 or 255, and |a _i -a ₀ |(i=1,2,...,k) is less than the set threshold, then use P _i '=a ₀ ±rand(X) to update the image The gray value of the block Pi ( _i =1,2,...,k), rand(X) represents a random number generated in the range of [1,X], and the updated image block is _Pi '.

4. The sonar image equalization method based on SLIC and adaptive filtering according to any one of claims 1-3, wherein the specific method of step 5 is: dividing the sonar image into a smaller number of image blocks.

5 . The sonar image equalization method based on SLIC and adaptive filtering according to claim 3 , wherein: X=5. 6 .

6 . The sonar image equalization method based on SLIC and adaptive filtering according to any one of claims 1 , 2 , 3 or 5 , wherein the number of image blocks is 3000. 7 .

7 . The sonar image equalization method based on SLIC and adaptive filtering according to claim 6 , wherein the number of image blocks in step 5 is 2500 and/or 2000 and/or 1500 and/or 1000. 8 .