CN110191345B - Incremental Lossless Compression Method for Separation of Foreground and Background Based on Huffman Coding - Google Patents

Abstract

The invention discloses a picture incremental lossless compression method based on Huffman coding separation of foreground and background. The invention uses the collection device to capture batch pictures of target objects in the same scene, adopts the lossless compression technology based on Huffman coding, and performs differential incremental compression on a large number of pictures captured in the same scene, and passes the foreground and background. The separation method makes the difference distribution more concentrated, so as to obtain the shortest possible coding length, obtain the best lossless compression effect, reduce the occupation of storage space, and can quickly decode and restore the picture for the subsequent upper layer use of the picture. The newly added pictures can also realize the above functions according to the threshold and the difference base information, which has great application value and economic significance.

Description

Incremental Lossless Compression Method for Separation of Foreground and Background Based on Huffman Coding

technical field

The invention belongs to the field of image compression, and relates to an incremental lossless compression method based on Huffman coding.

Background technique

In order to meet the daily operation inspection needs of power grid equipment, a substation of a power supply bureau of the State Grid has launched an intelligent inspection robot to take daily photos of the overlapping surfaces and instruments of the power grid to improve the efficiency of operation and maintenance. In the working scene, the inspection robot takes pictures of the key equipment of the substation and the overlapping surface along the established path, and identifies the readings of the relevant instruments and the temperature of the overlapping surface. However, the daily massive data of the robot includes three categories of visible light photos, infrared photos, and infrared data. The main image formats are jpg, bmp, png, etc., and the data generated by the robot are all stored in local ordinary PCs. Mass data provides the basis for data mining of power station operation and maintenance, but the large-scale, heterogeneous, and multi-source characteristics of data also bring huge challenges to data storage and use. Therefore, it is necessary to effectively compress massive images to reduce Storage pressure, and at the same time need to achieve lossless decompression so as not to lose information, to provide support for subsequent use.

Huffman coding is a widely used lossless compression method. It uses the probability of occurrence of characters to construct codewords with the shortest average length of different prefixes to achieve the purpose of shortest encoding length of all characters, which can effectively reduce the storage space of data. , has fewer restrictions, has strong versatility, and is easy to implement through programming.

SUMMARY OF THE INVENTION

The purpose of the present invention is to deal with the problem of storage pressure and inconvenience caused by the massive pictures generated by the intelligent inspection robot of the power grid substation, adopting the lossless compression technology based on Huffman coding, to make a difference between a large number of pictures taken in the same scene It can achieve the shortest possible encoding length, obtain the best lossless compression effect, reduce the storage space occupation, and realize rapid decoding. The image is restored for the subsequent use of the upper layer of the image. For the newly added image, the above functions can also be implemented according to the threshold and the difference base information, which has great application value and economic significance.

The object of the present invention is achieved by the following technical solutions: a picture incremental lossless compression method based on the separation of foreground and background of Huffman coding, the method comprises the following steps:

(1) The batch image group X of the target object in the same scene is captured by the acquisition device, and stored in the form of an array, which is expressed as:

Wherein, X _j is the three-dimensional matrix array storage form of the jth picture, and m is the total number of pictures taken by the acquisition device in the same scene;

Denote the storage space S of m pictures as:

Among them, S _j is the storage space before the jth picture is compressed;

(2) When compressing the picture, divide the original picture into three dimensions according to the R, G and B components of the picture for processing, and store them separately as three arrays of X _r , X _g , and X _b , Expressed as:

Wherein, X _rj is the array stored in the R component of the jth picture, X _gj is the array stored in the G component of the jth picture, and X _bj is the array stored in the B component of the jth picture;

(3) arbitrarily take Q pictures from the picture group taken in the same scene, take the average value of the pixel values at the same pixel position of the Q pictures, and use it as the pixel value of the pixel position of the base picture to obtain the base picture, and then use the The R, G, B components of all the pictures in the scene are differentiated with the base R, G, B components, respectively, to obtain the difference arrays ΔX _r , ΔX _g , ΔX _b , which can be expressed as:

Among them, ΔX _ri is the difference array between the R component of the ith picture and the base R component; ΔX _gi is the difference array of the G component of the ith picture and the base G component; ΔX _bi is the ith picture B component and the base B array of differences of components;

(4) Take the number of iterations k = 0, determine the threshold distribution interval [ε _min , ε _max ] of the foreground and background separation according to the distribution range of the difference array values, and select the initial threshold of the foreground and background separation ε _k = ε _min (for example, in 10 The initial threshold ε _k = 10) is selected in the -200 interval; according to the threshold ε _k , the part with pixel difference greater than or equal to ε _k is stored as the foreground part, and the part less than ε _k is stored as the background part, so that the picture array is divided into foreground and The two parts of the background are stored separately and recorded as ΔX _rq , ΔX _rb , ΔX _gq , ΔX _gb , ΔX _bq , ΔX _bb , which are expressed as:

ΔX _rq =[ΔX _rq1 ,ΔX _rq2 ,...,ΔX _rqm ] ^T

ΔX _rb =[ΔX _rb1 ,ΔX _rb2 ,...,ΔX _rbm ] ^T

ΔX _gq =[ΔX _gq1 ,ΔX _gq2 ,...,ΔX _gqm ] ^T

ΔX _gb =[ΔX _gb1 ,ΔX _gb2 ,...,ΔX _gbm ] ^T

ΔX _bq =[ΔX _bq1 ,ΔX _bq2 ,...,ΔX _bqm ] ^T

ΔX _bb =[ΔX _bb1 ,ΔX _bb2 ,...,ΔX _bbm ] ^T

Among them, ΔX _rqi is the difference array between the foreground red component and the base red component of the ith picture; ΔX _rbi is the difference array of the background red component and the base red component of the ith picture; ΔX _gqi is the foreground green component of the ith picture Difference array with the base green component; ΔX _gbi is the difference array between the background green component and base green component of the ith picture; ΔX _bqi is the difference array between the foreground blue component and the base blue component of the ith picture; ΔX _bbi is the difference array between the background blue component and the base blue component of the ith picture;

表示为：(5) After obtaining the above R, G, B difference arrays with relatively concentrated distribution, use classical Huffman coding to compress and store ΔX _rqi , ΔX _rbi , ΔX _gqi , ΔX _gbi , ΔX _bqi , ΔX _bbi losslessly, and obtain the average Code length

Expressed as:

为第i张图片前景红色分量与基底红色分量的差值数组的平均编码长度，

张图片前景红色分量与基底红色分量的差值数组第a个像素点的编码长度，n_rqi为第i张图片前景红色分量与基底红色分量的差值数组像素点的个数，其他符号含义可以此类推，为避免赘述，此处不再详细说明；in,

is the average encoding length of the difference array between the foreground red component and the base red component of the ith picture,

The coding length of the a-th pixel of the difference array between the foreground red component and the base red component of the picture, n _rqi is the number of pixels in the difference array between the foreground red component and the base red component of the ith picture, and the meanings of other symbols can be And so on, in order to avoid redundant description, it will not be described in detail here;

(6) After obtaining the average coding length of the three dimensions of R, G and B of the foreground and background, calculate the lossless compression ratios C _ri , C _gi , C _bi of the three components of R, G and B of the i-th picture, and the expressions are as follows :

Wherein, n _ri is the total number of R component pixels of the ith picture; n _gi is the total number of G component pixels of the ith picture; n _bi is the total number of B component pixels of the ith picture, and n _ri =n _gi =n _bi =n _i /3, which are equal to 1/3 of the total number of pixels n _i in the i-th picture.

(7) The lossless compression ratio of the ith picture is C _i , which can be expressed as:

When the total number of pictures to be compressed is n, the total lossless compression ratio C ₀ is:

(8) When ε _k =ε _max , go to step (9); when ε _k <ε _max , k=k+1, ε _k =ε _k-1 +Δε, Δε is the threshold step size (optional 10), repeat step (4) to step (7), obtain new total lossless compression ratio C _k ;

(9) Select the threshold corresponding to the minimum total lossless compression ratio as the final threshold for picture compression in this scene to perform lossless compression processing of batch pictures; after picture i is lossless compressed, it is stored with the pixel point difference value of the foreground and background in three dimensions Code for six groups of Huffman;

(10) When restoring picture i, restore the foreground and background codes of each dimension to the pixels before compression, and then sum with the base pixels, and display the three dimensions together to obtain the original picture before lossless compression , according to this principle, the lossless restoration of batches of compressed pictures can be achieved to support subsequent upper-layer use.

Further, the method is applied to a substation intelligent inspection robot in the same scene to shoot a batch of pictures of the target object at fixed points along a fixed route, and compress the pictures. Because the inspection robot takes pictures at fixed points along a fixed route, the background (sky, etc.) and the foreground (telephone poles, etc.) are obviously different, so that after separation according to the threshold, the difference distribution is more concentrated than before separation.

The beneficial effects of the present invention are: the present invention adopts the lossless compression technology based on Huffman coding, performs differential incremental compression on a large number of pictures taken in the same scene, and separates the foreground and the background to make the difference distribution more centralized to obtain the shortest encoding length as possible, obtain the best lossless compression effect, reduce the storage space occupation, and can quickly decode and restore the picture for the subsequent upper layer of the picture. The threshold value and the difference base information realize the above functions, and have great application value and economic significance.

Description of drawings

Fig. 1 is the flow chart of the incremental lossless compression method of the present invention based on Huffman coding foreground, background separation;

Figure 2 is a visual picture of the R component difference array of a certain picture;

Figure 3 is a visual image of the difference array with the foreground and background separated by thresholding.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

In order to meet the needs of intelligent inspection, power companies continue to increase investment in inspection robots in daily substation operations, replacing manual inspections to a certain extent. Robot inspections generate a large number of pictures every day, and the current storage technology cannot cope with large-scale picture increments. Therefore, a lot of money will be needed to expand storage equipment for data, pictures and other resources in the future. The research of obtaining pictures in the same scene found that the shooting background is basically the same, and the shooting foreground is relatively different. By extracting the common background (ie, the base image), and from each pixel, the difference between the picture and the common background is calculated, and the difference is divided into blocks Obviously, the present invention adopts a lossless incremental image compression technology based on Huffman coding based on the separation of foreground and background according to the feature that the inspection robot takes pictures at fixed points along a fixed route every day, so that the similarity of the shooting content in the same scene is high. , reduce the storage pressure of massive pictures, realize the unified storage management of lossless compressed image data, and provide support for compressed storage, rapid retrieval and transmission, and intelligent analysis of large-scale application of robot data in the future.

An incremental lossless compression method based on Huffman coding for the separation of foreground and background proposed by the present invention can be used to compress a class of pictures with a common background, as shown in FIG. 1 , including the following steps:

(1) In this example, a substation in Wenzhou Power Grid is taken as an example, and the .bmp format pictures (both 240*320 in size) taken within one month of the robot inspection tower wiring shooting scene are selected for lossless compression processing, and MATLAB is used to read the pictures Then, take a .bmp image as an example to get a three-dimensional array of 240*320*3:

(2) During the compression process, the original image is divided into three dimensions of R, G, B, and processed separately, and stored separately as three arrays of X _r , X _g , and X _b , which can be respectively obtained with a size of 240* Three arrays of 320:

(3) Taking the collection of 10 pictures as an example, record the storage space S of the 10 pictures before compression as (unit: bit):

(4) Take any 20 pictures from the shooting pictures of the same scene, and take the average value of each pixel point as the base, and compare the R, G, B components of other pictures in this scene with the base R, G, B components respectively. Make a difference to get the difference arrays ΔX _r , ΔX _g , ΔX _b , take a picture to be compressed as an example, the visualization of ΔX _r is shown in Figure 2 below:

(5) According to the numerical distribution of the difference array, take 10 as the step, select and determine the optimal threshold between 0 and 200, so as to minimize the compression ratio, and store the difference between the foreground and the background separately, and record them as ΔX _rq , ΔX _rb , ΔX _gq ΔX _gb , ΔX _bq , ΔX _bb , the threshold here is 50, taking a picture to be compressed as an example, the foreground and background of which are separated ΔX _rq , ΔX _rb are visualized as shown in Figure 3 below .

其中一张待压缩图片的各平均编码长度如下表1所示：(6) After obtaining the above R, G, B difference arrays with relatively concentrated distribution, use classical Huffman coding to perform lossless compression and storage on ΔX _rqi , ΔX _rbi , ΔX _gqi , ΔX _gbi , ΔX _bqi , ΔX _bbi , we can get average code length

The average encoding length of one of the pictures to be compressed is shown in Table 1 below:

Table 1

(7) After obtaining the average coding length of the three dimensions of R, G, and B of the foreground and background, calculate the lossless compression ratios C _ri , C _gi , and C _bi of the three components of R, G, and B of the ith picture, and use one of them Taking a picture to be compressed as an example, the compression ratio of its R, G, and B components is shown in Table 2 below:

Table 2

(8) The lossless compression ratio of the picture is C:

When the total number of pictures to be compressed is n=10, the total lossless compression ratio C is:

(9) According to the corresponding decoding technology, after summing the to-be-decompressed picture and the base, decompression can be realized, that is, fast and lossless restoration of batch compressed pictures can be realized to support subsequent upper-layer use.

After the method of the invention is tested on the non-public large-scale test set provided by Wenzhou Power Grid Corporation, the compression ratio is selected according to the base image and the separation threshold, and the range of the lossless compression ratio is distributed in the closed interval [0.1764, 0.1985], and the compression effect is remarkable. It can effectively save storage space and has great economic significance and application value.

The above-mentioned embodiments are used to explain the present invention, rather than limit the present invention. Within the spirit of the present invention and the protection scope of the claims, any modifications and changes made to the present invention all fall into the protection scope of the present invention.

Claims (2)

1. a picture incremental lossless compression method based on the foreground of Huffman coding, background separation, is characterized in that, comprises the following steps: (1) The batch image group X of the target object in the same scene is captured by the acquisition device, and stored in the form of an array, which is expressed as:

Wherein, X _j is the three-dimensional matrix array storage form of the jth picture, and m is the total number of pictures taken by the acquisition device in the same scene; Denote the storage space S of m pictures as:

Among them, S _j is the storage space before the jth picture is compressed; (2) When compressing the picture, divide the original picture into three dimensions according to the R, G and B components of the picture for processing, and store them separately as three arrays of X _r , X _g , and X _b , Expressed as:

Wherein, X _rj is the array stored in the R component of the jth picture, X _gj is the array stored in the G component of the jth picture, and X _bj is the array stored in the B component of the jth picture; (3) arbitrarily take Q pictures from the picture group taken in the same scene, take the average value of the pixel values at the same pixel position of the Q pictures, and use it as the pixel value of the pixel position of the base picture to obtain the base picture, and then use the The R, G, B components of all the pictures in the scene are differentiated with the base R, G, B components, respectively, to obtain the difference arrays ΔX _r , ΔX _g , ΔX _b , which can be expressed as:

Among them, ΔX _ri is the difference array between the R component of the ith picture and the base R component; ΔX _gi is the difference array of the G component of the ith picture and the base G component; ΔX _bi is the ith picture B component and the base B array of differences of components; (4) Take the number of iterations k=0, determine the threshold distribution interval [ε _min , ε _max ] of the foreground and background separation according to the distribution range of the difference array values, and select the initial threshold ε _k =ε _min of the foreground and background separation; according to the threshold ε _k , the part with pixel difference greater than or equal to ε _k is stored as the foreground part, and the part less than ε _k is stored as the background part, so the image array is divided into two parts, the foreground and the background, which are stored separately, and are recorded as ΔX _rq , ΔX _rb , ΔX _gq , ΔX _gb , ΔX _bq , ΔX _bb , expressed as: ΔX _rq =[ΔX _rq1 ,ΔX _rq2 ,...,ΔX _rqm ] ^T ΔX _rb =[ΔX _rb1 ,ΔX _rb2 ,...,ΔX _rbm ] ^T ΔX _gq =[ΔX _gq1 ,ΔX _gq2 ,...,ΔX _gqm ] ^T ΔX _gb =[ΔX _gb1 ,ΔX _gb2 ,...,ΔX _gbm ] ^T ΔX _bq =[ΔX _bq1 ,ΔX _bq2 ,...,ΔX _bqm ] ^T ΔX _bb =[ΔX _bb1 ,ΔX _bb2 ,...,ΔX _bbm ] ^T Among them, ΔX _rqi is the difference array between the foreground red component and the base red component of the ith picture; ΔX _rbi is the difference array of the background red component and the base red component of the ith picture; ΔX _gqi is the foreground green component of the ith picture Difference array with the base green component; ΔX _gbi is the difference array between the background green component and base green component of the ith picture; ΔX _bqi is the difference array between the foreground blue component and the base blue component of the ith picture; ΔX _bbi is the difference array between the background blue component and the base blue component of the ith picture; (5) After obtaining the above R, G, B difference arrays with relatively concentrated distribution, use classical Huffman coding to compress and store ΔX _rqi , ΔX _rbi , ΔX _gqi , ΔX _gbi , ΔX _bqi , ΔX _bbi losslessly, and obtain the average Code length

Expressed as:

in,

is the average encoding length of the difference array between the foreground red component and the base red component of the ith picture,

is the coding length of the a-th pixel of the difference array between the foreground red component and the base red component of the ith picture, and n _rqi is the number of pixels in the difference array between the foreground red component and the base red component of the ith picture;

is the average encoding length of the difference array between the background red component and the base red component of the ith picture;

is the average coding length of the difference array between the foreground green component and the base green component of the ith picture;

is the average coding length of the difference array between the background green component and the base green component of the ith picture;

is the average encoding length of the difference array between the foreground blue component and the base blue component of the ith picture;

is the average encoding length of the difference array between the background blue component and the base blue component of the ith picture;

The encoding length of the a-th pixel of the difference array between the foreground green component and the base green component of the i-th picture,

is the coding length of the a-th pixel of the difference array between the foreground blue component and the base blue component of the i-th picture,

is the coding length of the a-th pixel of the difference array between the background red component and the base red component of the i-th picture,

is the coding length of the a-th pixel of the difference array between the background green component and the base green component of the i-th picture,

is the coding length of the a-th pixel of the difference array between the background blue component and the base blue component of the ith picture; n _gq i The number of pixels of the difference array between the foreground green component and the base green component of the ith picture , n _bqi is the number of pixels in the difference array between the foreground blue component and the base blue component of the ith picture, and n _rbi is the number of pixels in the difference array between the background red component and the base red component of the ith picture , n _gbi is the number of pixels in the difference array between the background green component and the base green component of the ith picture, n _bb i is the number of pixels in the difference array between the background blue component and the base blue component of the ith picture number; (6) After obtaining the average coding length of the three dimensions of R, G and B of the foreground and background, calculate the lossless compression ratios C _ri , C _gi , C _bi of the three components of R, G and B of the i-th picture, and the expressions are as follows :

Wherein, n _ri is the total number of R component pixels of the ith picture; n _gi is the total number of G component pixels of the ith picture; n _bi is the total number of B component pixels of the ith picture, and n _ri =n _gi =n _bi =n _i /3, which are equal to 1/3 of the total number of pixels n _i of the i-th picture; (7) The lossless compression ratio of the ith picture is C _i , which can be expressed as:

When the total number of pictures to be compressed is n, the total lossless compression ratio C ₀ is:

(8) When ε _k =ε _max , go to step (9); when ε _k <ε _max , k=k+1, ε _k =ε _k-1 +Δε, Δε is the threshold step size, repeat step ( 4) to step (7), obtain a new total lossless compression ratio C _k ; (9) Select the threshold corresponding to the minimum total lossless compression ratio as the final threshold for picture compression in this scene to perform lossless compression processing of batch pictures; after picture i is lossless compressed, it is stored with the pixel point difference value of the foreground and background in three dimensions Code for six groups of Huffman; (10) When restoring picture i, restore the foreground and background codes of each dimension to the pixels before compression, and then sum with the base pixels, and display the three dimensions together to obtain the original picture before lossless compression . 2. a kind of picture incremental lossless compression method based on the foreground of Huffman coding, background separation according to claim 1, is characterized in that, this method is applied to a certain substation intelligent inspection robot in the same scene along the fixed edge The route fixed-point shooting batch picture group of the target object, and the picture is compressed.

Images