CN105550690B - A kind of optimization method based on Stentiford vision modes - Google Patents
A kind of optimization method based on Stentiford vision modes Download PDFInfo
- Publication number
- CN105550690B CN105550690B CN201510926402.9A CN201510926402A CN105550690B CN 105550690 B CN105550690 B CN 105550690B CN 201510926402 A CN201510926402 A CN 201510926402A CN 105550690 B CN105550690 B CN 105550690B
- Authority
- CN
- China
- Prior art keywords
- pixel
- value
- gray level
- level image
- pending gray
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Abstract
The present invention relates to a kind of optimization methods based on Stentiford vision modes, design introduces Uniform LBP algorithms and statistics with histogram method, it is excessive to solve traditional Stentiford vision mode model randomnesss, it is excessively sensitive to details, calculate time-consuming disadvantage, and experiments have shown that, method designed by the present invention can greatly improve the effect of image-region extraction, and processing speed is substantially increased, image processing efficiency is obviously improved, and has stronger practical value.
Description
Technical field
The present invention relates to a kind of optimization methods based on Stentiford vision modes, belong to technical field of image processing.
Background technology
In recent years, with the rapid development of multimedia technology and computer network, the capacity sharp increase of digital picture.Only
Corresponding algorithm is run for interesting image regions, the scale of data on the one hand can be reduced, to improve operational efficiency；It is another
Aspect can also reduce non-ROI and be interfered caused by result.Therefore the extractive technique of interesting image regions has become grinds at present
The hot issue studied carefully.
Have much currently based on region of interesting extraction (region of interest abbreviation ROI) method of image, but
Four classes can be divided by being macromethod.
(1) method being manually specified, such as (B Moghaddam, H Biermann, D Margaritis.Defining
image content with multiple regions-of-interest[J].IEEE Workshop on Content-
Based Access of Image and Video Libraries, 1999.), it is manual according to the knowledge of oneself by user
The advantages of specified region, such method be machine can accurate understanding user true intention, but due to excessive artificial of process
It participates in, so interactive process is unfriendly.
(2) human eye viewpoint back tracking method, such as (SR Research Ltd.EyeLink II user manual version
2.12[R].Canada:SRResearch Ltd., 2006.) it is to be regarded when capturing eye-observation picture by special instrument and equipment
Point position, record fixation time watch the indexs such as coordinate points and eye blink response attentively, and founding mathematical models obtain comparison conspicuousness water
Flat, the ROI to tentatively obtain, advantage is can to react observer well to pay attention to region, but due to needing profession in the process
Equipment haves the characteristics that practicality is poor for user's application.
(3) dividing method of special object, such as (LItti, C Koch, E Niebur.Amodel of saliency
based visual attention for rapid scene analysis[J].IEEE Transactions on
Pattern Analysis and Machine Intelligence,1998,20(11):1254-1259.) include traditional figure
As dividing method.For example, dividing ridge method, small wave converting method etc., these method specific aims are stronger, only for certain spies
It is very ideal to determine image effect.Thus, it can be known that the method conceals a premise, if i.e. special object area-of-interest.
(4) visual attention model method, such as (F W M Stentiford.An attention based similarity
measure with application to content based information retrieval[C]
.Proceedings of the Storage and Retrieval for Media Databases
Conference.Bellingham:Society of Photo-Optical Instrume ntation Engineers,
2003:221-232.) it is the model simulated human-eye visual characteristic and established, builds visual saliency map.Wherein, two famous moulds
Type is Itti vision modes and Stentiford models.Itti vision modes change the reason of image effect greatly for object color
Think.But there is the image effect of anomalistic point poor in the uniform image of, color larger for obvious object.Stentiford model ratios
Itti vision mode effects are relatively sharp and accurate.However Stentiford models have some disadvantages to restrict its application range.
First, it is to be randomly selected to certain neighborhood of pixels structure in algorithm, this makes result carry excessive randomness.Secondly, algorithm pair
It is excessively sensitive in details, it causes to regard target area at bulk zone in non-uniform background.Finally, computationally intensive, seriously account for
With resource, take considerable time.
Invention content
In view of the above technical problems, technical problem to be solved by the invention is to provide one kind being based on Stentiford visions
The optimization method of model, design introduce Uniform LBP algorithms and solve big and excessively sensitive to the details disadvantage of randomness, can
The effect of image-region extraction is greatly improved, and statistics with histogram method is introduced by the design of three-dimensional array, is carried significantly
High processing speed, image processing efficiency are obviously improved.
In order to solve the above-mentioned technical problem the present invention uses following technical scheme：The present invention devises one kind and is based on
The optimization method of Stentiford vision modes, includes the following steps：
Step 001. carries out gray proces for pending coloured image, obtains pending gray level image, and enter step
002；
Each pixel that step 002. is directed in pending gray level image respectively sets the picture centered on pixel
Round measuring and calculating region corresponding to vegetarian refreshments, wherein the radius that the circle calculates region is the distance between neighbor pixel, the circle
It includes the pixel and up and down four pixel adjacent with the pixel that shape, which calculates region,；Thus to obtain pending ash
The corresponding round measuring and calculating region of each pixel difference in image is spent, subsequently into step 003；
Step 003. is directed to each pixel in pending gray level image respectively, by round measuring and calculating area corresponding to pixel
Up and down four pixel adjacent with the pixel in domain, and on the circumference that the circle calculates region, adjacent picture
Eight objects adjacent with the pixel in region are calculated in centre position between vegetarian refreshments as round corresponding to the pixel,
The pixel value for obtaining eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel, subsequently into step
004；
Step 004. is directed to each pixel in pending gray level image respectively, calculates for round corresponding to pixel
Whether eight objects adjacent with the pixel in region, the pixel value for being respectively compared each object are more than or equal to the pixel
Pixel value, it is then 1 for the object tag to be, it is 0 to be otherwise directed to the object tag；Then it is directed to the label of eight objects
Value is combined by default initial position and sequence and constitutes an eight bit, and the eight bit is converted to
Decimal value corresponds to the pixel；Thus each pixel difference is corresponding in pending gray level image one is obtained
A decimal value, enters step 005；
Step 005. will be each in pending gray level image using Uniform LBP patterns according to pending gray level image
Pixel corresponding decimal value respectively, be converted to it is corresponding with pending gray level image, include 59 kinds of numerical value
Uniform LBP matrixes, wherein of the number of element and pixel in pending gray level image in Uniform LBP matrixes
Number is equal, and corresponds, and the value of each element is the grade point as corresponding pixel points, Uniform in Uniform LBP matrixes
A total of 59 kinds of numerical value of the value of element in LBP matrixes, i.e., a total of 59 kinds of the grade point of pixel in pending gray level image；So
After enter step 006；
Uniform LBP matrix of the step 006. corresponding to pending gray level image, for pending gray level image
In each pixel, establish the three-dimensional array [Rank, Value, Location (x, y)] corresponding to pixel respectively, wherein
Rank indicates that the grade point of corresponding pixel points, Value indicate that the pixel value of corresponding pixel points, Location (x, y) indicate to correspond to
The coordinate of pixel；Subsequently into step 007；
Three-dimensional array of the step 007. corresponding to each pixel in pending gray level image, by pending gray-scale map
The grade point Rank that all pixels as in press corresponding to it is divided, and is carried out statistics with histogram, that is, is obtained each grade point
The corresponding pixel collection R of Rank institutes*, subsequently into step 008；
Step 008. be directed to 59 kinds of grade point Rank, by minimum level value Rank, by grade point Rank from as low as greatly
Sequence, sequentially respectively be directed to each grade point Rank corresponding to pixel collection R*, for pixel collection R*In it is each
Pixel 00801 is operated to step 00803, and then obtains each picture in pending gray level image as follows respectively
The significance of vegetarian refreshments；
Step 00801. pixel collection R where pixel*In in other each pixels in addition to the pixel,
Arbitrary take out presets N number of pixel, the set T corresponding to the pixel is formed, subsequently into step 00802；Wherein, N is less than
Equal to each pixel collection R*In minimum pixel number subtract 1；
Step 00802. is directed to the pixel, counts unequal with pixel pixel value Value in its correspondence set T
The number P of pixel, and enter step 00803；
Step 00803. calculatesObtain the significance of the pixel.
As a preferred technical solution of the present invention：After the step 008, further includes step 009, execute the step
After 008,009 is entered step, wherein step 009 is as follows：
Step 009. is directed to the significance of each pixel in pending gray level image, is multiplied by preset ratio coefficient respectively,
The significance for updating each pixel in pending gray level image obtains corresponding notable figure further according to pending gray level image
Picture.
As a preferred technical solution of the present invention：The preset ratio coefficient is 255.
As a preferred technical solution of the present invention：The step 003 specifically includes as follows：
The each pixel being directed to respectively in pending gray level image, according to adjacent four pixels up and down of pixel
Pixel value, and the upper left adjacent with the pixel, upper right, lower-left, four pixels in bottom right pixel value, be somebody's turn to do respectively
Circle measuring and calculating region circumference on, between neighbor pixel centre position pixel value；It then will be round corresponding to the pixel
Calculate up and down four pixels adjacent with the pixel and circle in region calculate on the circumference in region, is adjacent
Centre position between pixel, it is right as adjacent with the pixel in round measuring and calculating region corresponding to the pixel eight
As obtaining the pixel value of eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel.
As a preferred technical solution of the present invention：In the step 003, it is directed in pending gray level image respectively
Each pixel, according to the pixel value of adjacent four pixels up and down of pixel, and the upper left adjacent with the pixel,
Upper right, lower-left, four pixels in bottom right pixel value obtain circle measuring and calculating area respectively using four bilinear interpolation methods
On the circumference in domain, between neighbor pixel centre position pixel value.
A kind of optimization method based on Stentiford vision modes of the present invention using above technical scheme with it is existing
Technology is compared, and is had the following technical effects：A kind of optimization method based on Stentiford vision modes that the present invention designs, if
Meter introduces the statistics with histogram method of Uniform LBP algorithms and three-dimensional array formula, and it is big and excessively quick to details to solve randomness
The shortcomings that sense, and it is demonstrated experimentally that method designed by the present invention can greatly improve the effect of image-region extraction, and significantly
Processing speed is improved, image processing efficiency is obviously improved.
Description of the drawings
Fig. 1 is that the present invention designs a kind of flow diagram of the optimization method based on Stentiford vision modes；
Fig. 2 is that the application of step 003 in a kind of optimization method based on Stentiford vision modes of present invention design is shown
It is intended to；
Fig. 3 a are the design sketch of the application tradition Stentiford methods of embodiment one；
Fig. 3 b are the design sketch of the artificial segmentation image method of the application of embodiment one；
Fig. 3 c are the design sketch of improved Itti vision modes method in the application document of embodiment one [18]；
Fig. 3 d are the design sketch of the GBVS vision mode methods in the application document of embodiment one [19]；
Fig. 3 e are the design sketch of the remaining model method of spectrum in the application document of embodiment one [20]；
Fig. 3 f are the design sketch of method designed by the application present invention of embodiment one；
Fig. 4 a are the design sketch of the application tradition Stentiford methods of embodiment two；
Fig. 4 b are the design sketch of the artificial segmentation image method of the application of embodiment two；
Fig. 4 c are the design sketch of improved Itti vision modes method in the application document of embodiment two [18]；
Fig. 4 d are the design sketch of the GBVS vision mode methods in the application document of embodiment two [19]；
Fig. 4 e are the design sketch of the remaining model method of spectrum in the application document of embodiment two [20]；
Fig. 4 f are the design sketch of method designed by the application present invention of embodiment two；
Fig. 5 a are the design sketch of the application tradition Stentiford methods of embodiment three；
Fig. 5 b are the design sketch of the artificial segmentation image method of the application of embodiment three；
Fig. 5 c are the design sketch of improved Itti vision modes method in the application document of embodiment three [18]；
Fig. 5 d are the design sketch of the GBVS vision mode methods in the application document of embodiment three [19]；
Fig. 5 e are the design sketch of the remaining model method of spectrum in the application document of embodiment three [20]；
Fig. 5 f are the design sketch of method designed by the application present invention of embodiment three；
Fig. 6 a are the design sketch of example IV application tradition Stentiford methods；
Fig. 6 b are the design sketch of the artificial segmentation image method of example IV application；
Fig. 6 c are the design sketch of improved Itti vision modes method in example IV application document [18]；
Fig. 6 d are the design sketch of the GBVS vision mode methods in example IV application document [19]；
Fig. 6 e are the design sketch of the remaining model method of spectrum in example IV application document [20]；
Fig. 6 f are the design sketch of method designed by the example IV application present invention.
Specific implementation mode
Specific embodiments of the present invention will be described in further detail for needle with reference to the accompanying drawings of the specification.
As shown in Figure 1, a kind of optimization method based on Stentiford vision modes that the present invention designs is in practical application
In process, specifically comprise the following steps：
Step 001. carries out gray proces for pending coloured image, obtains pending gray level image, and enter step
002。
Each pixel that step 002. is directed in pending gray level image respectively sets the picture centered on pixel
Round measuring and calculating region corresponding to vegetarian refreshments, wherein the radius that the circle calculates region is the distance between neighbor pixel, the circle
It includes the pixel and up and down four pixel adjacent with the pixel that shape, which calculates region,；Thus to obtain pending ash
The corresponding round measuring and calculating region of each pixel difference in image is spent, subsequently into step 003.
Step 003. is directed to each pixel in pending gray level image respectively, according to pixel adjacent up and down four
The pixel value of a pixel, and the upper left adjacent with the pixel, upper right, lower-left, four pixels in bottom right pixel value, adopt
With four bilinear interpolation methods, obtain respectively on the circumference in circle measuring and calculating region, centre position between neighbor pixel
Pixel value；I.e. as shown in Figure 2, wherein a, b, c, d are four pixels, and 2,4,6,8 be round measuring and calculating area corresponding to pixel c
Centre position on the circumference in domain between tetra- pixel neighbor pixels of a, b, c, d, according to fa、fb、fc、fdRespectively obtain a,
B, the pixel value of tetra- pixels of c, d, then passes through following the separate equations：
f2≈0.49×fa+0.21×fb+0.09×fc+0.21×fd
f4≈0.49×fb+0.21×fa+0.09×fd+0.21×fc
f6≈0.49×fc+0.21×fb+0.09×fa+0.21×fd
f8≈0.49×fd+0.21×fa+0.09×fd+0.21×fc
The pixel value f of 2,4,6,8 positions is obtained respectively2、f4、f6、f8, then by round measuring and calculating area corresponding to the pixel
In domain up and down four pixels adjacent with the pixel and the circle measuring and calculating region circumference on, neighbor pixel
Between centre position obtained as eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel
The pixel value for obtaining eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel, subsequently into step 004.
Step 004. is directed to each pixel in pending gray level image respectively, calculates for round corresponding to pixel
Whether eight objects adjacent with the pixel in region, the pixel value for being respectively compared each object are more than or equal to the pixel
Pixel value, it is then 1 for the object tag to be, it is 0 to be otherwise directed to the object tag；Then it is directed to the label of eight objects
Value is combined by default initial position and sequence and constitutes an eight bit, and the eight bit is converted to
Decimal value corresponds to the pixel；Thus each pixel difference is corresponding in pending gray level image one is obtained
A decimal value, enters step 005.
Step 005. will be each in pending gray level image using Uniform LBP patterns according to pending gray level image
Pixel corresponding decimal value respectively, be converted to it is corresponding with pending gray level image, include 59 kinds of numerical value
Uniform LBP matrixes, wherein of the number of element and pixel in pending gray level image in Uniform LBP matrixes
Number is equal, and corresponds, and the value of each element is the grade point as corresponding pixel points, Uniform in Uniform LBP matrixes
A total of 59 kinds of numerical value of the value of element in LBP matrixes, i.e., a total of 59 kinds of the grade point of pixel in pending gray level image；So
After enter step 006.
Uniform LBP matrix of the step 006. corresponding to pending gray level image, for pending gray level image
In each pixel, establish the three-dimensional array [Rank, Value, Location (x, y)] corresponding to pixel respectively, wherein
Rank indicates that the grade point of corresponding pixel points, Value indicate that the pixel value of corresponding pixel points, Location (x, y) indicate to correspond to
The coordinate of pixel；Subsequently into step 007.
Three-dimensional array of the step 007. corresponding to each pixel in pending gray level image, by pending gray-scale map
The grade point Rank that all pixels as in press corresponding to it is divided, and is carried out statistics with histogram, that is, is obtained each grade point
The corresponding pixel collection R of Rank institutes*, subsequently into step 008.
Step 008. be directed to 59 kinds of grade point Rank, by minimum level value Rank, by grade point Rank from as low as greatly
Sequence, sequentially respectively be directed to each grade point Rank corresponding to pixel collection R*, for pixel collection R*In it is each
Pixel 00801 is operated to step 00803, and then obtains each picture in pending gray level image as follows respectively
The significance of vegetarian refreshments, subsequently into step 009.
Step 00801. pixel collection R where pixel*In in other each pixels in addition to the pixel,
Arbitrary take out presets N number of pixel, the set T corresponding to the pixel is formed, subsequently into step 00802；Wherein, N is less than
Equal to each pixel collection R*In minimum pixel number subtract 1.
Step 00802. is directed to the pixel, counts unequal with pixel pixel value Value in its correspondence set T
The number P of pixel, and enter step 00803.
Step 00803. calculatesObtain the significance of the pixel.
Step 009. is directed to the significance of each pixel in pending gray level image, is multiplied by preset ratio coefficient respectively
255, the significance of each pixel in pending gray level image is updated, further according to pending gray level image, is obtained corresponding aobvious
Write image.
Practical application is carried out based on the above-mentioned specific designed optimization method based on Stentiford vision modes, is such as schemed
3a -3f, 4a -4f, 5a -5f, shown in 6a -6f, and divide respectively with traditional Stentiford methods, manually image method,
Document [18] (Song Kan are based on marking area area detection [J] the computer technologies for improving visual attention model and develop, 2015,
7th phase (07):(peak Wang Xing, Shao Zhen view-based access control models are notable for improved Itti vision modes method, document [19] in 234-236)
Remote Sensing Image Retrieval method [J] Surveying and mappings of point feature, 2014,39 (04):GBVS vision modes method, text in 34-38)
Offer [20] (Yin Chunxia, Xu De, Li Chengrong, wait based on notable figure SIFT feature detection and match [J] computer engineering,
2012,38(16):The remaining model method application of spectrum is compared in 189-191), wherein Fig. 3 a, 4a, 5a, 6a are tradition
Stendiford vision modes generate；Fig. 3 b, 4b, 5b, 6b are that people's work point cuts image method, and image target area is set as 1,
Set background to 0. bianry image；Fig. 3 c -3e, 4c -4e, 5c -5e, 6c -6e are respectively to changing in document [18]
Into Itti vision modes, the obtained notable figure of the GBVS vision modes in document [19], the remaining model of the spectrum in document [20]
On the basis of, by specific image gray scale interval by [0,1], it is transformed into [0,255] in proportion and converts obtained image, from
And final notable figure is the gray level image in [0,255] section, target area is black, and region shows that significance is higher more deeply feeling；Figure
3f, 4f, 5f, 6f are after being calculated by design method of the present invention, and denoising generates.
Observation chart 3a -3f, 4a -4f, 5a -5f, shown in 6a -6f, it is known that design method effect of the present invention is integrally notable
Better than other methods.The comparison of design method of the present invention and traditional Stentiford methods, shows improved method, can be with
Randomness is reduced, is preferably solved to details excessively sensitive issue, for quantitative all kinds of algorithm process effects of comparison, this hair
(Ma Runing, applies clivia to bright reference literature [21], and Ding Jundi waits vision significances to highlight evaluation [J] robotics of target
Report, 2012,38 (5):870-876), it is calculated.
Image Precision, Recall and F-measure are calculated separately according to following formula (to be denoted as respectively for the sake of simple
P,R,F)
P=∑s ((1-S) × B)/∑ (1-S)
R=∑s ((1-S) × B)/∑ B
F=2 × P × R/ (P+R)
Wherein S is the notable figure that (0,255) gray scale notable figure that all kinds of algorithms generate is transformed into (0,1) section in proportion；B
For people's work segmentation figure；∑ represents the summation of all pixels point gray value；(1-S) × B indicates what two image pixel point values were multiplied
Gray-scale map；It is bigger to can be understood as F values, notable figure display target effect is more ideal.All kinds of algorithms are by calculating the every width of database
Precision, Recall, F-measure of image are counted and are averaged.
For Precision indexs, GBVS vision mode methods are better than document in design method of the present invention, document [19]
[18] the remaining model method of the middle spectrum of improvement Itti vision modes method, document [20], tradition Stentiford methods in, and this hair
Bright design method is better than GBVS vision mode methods again.For Recall indexs, design method of the present invention, tradition
Better than being composed in GBVS vision modes method, document [20], remaining model method, document [18] are middle to be improved Stentiford methods
Itti vision mode methods, and design method of the present invention is better than tradition Stentiford methods.Illustrate traditional Stentiford
Method itself there is certain sensibility, extraction image significance to have some superiority details, and target area can be made higher aobvious
Work degree, but traditional method is excessively sensitive to details, and the present invention greatly improves extraction effect after improving so that extraction effect
Clearly, accurately.Moreover, compared with other three kinds of methods, extraction effect is satisfactory.
About traditional Stentiford methods and design method of the present invention, the links time is multiple in handling image process
Miscellaneous degree is as shown in table 1 below：
| Conventional method | The method of the present invention | |
| Interpolation | O[T] | |
| LBP matrixes | O[T] | |
| Statistics with histogram | O[T] | |
| Significance | O[T×T] | O[T] |
| Notable figure | O[T] | O[T] |
Table 1
Wherein, during calculating significance, because to there is the process for finding each element same pixel set, then exist
Take N points in set at random, T is the pixel number in image, and O [] is complexity function, thus, it can be known that conventional method when
Between complexity be O [T × T], and the time complexity of the designed optimization method based on Stentiford vision modes of the present invention
For O [T], it can be seen that, the designed optimization method based on Stentiford vision modes of the present invention is shown relative to conventional method
Work improves image processing efficiency.
The improved vision mode sides Itti in tradition Stentiford methods, design method of the present invention, document [18] as a result,
The time that the remaining model method application of spectrum is compared in GBVS vision modes method, document [20] in method, document [19] is complicated
Degree is as shown in table 2 below
| Model classification | Time complexity |
| Improved Itti vision modes method in document [18] | O[T×Log2T] |
| GBVS vision mode methods in document [19] | O[T×Log2T] |
| The remaining model method of spectrum in document [20] | O[T×Log2T] |
| Traditional Stentiford methods | O[T×T] |
| Design method of the present invention | O[T] |
Table 2
In summary：The method of the present invention is efficient not only than traditional Stendiford methods significant effect, and relative to
The GBVS vision modes type method in Itti vision modes type method, document [19] in document [18], in document [20]
Spectrum residue hypothesis types of models method also have certain advantage.
It is explained in detail for embodiments of the present invention above in conjunction with Figure of description, but the present invention is not limited to
The above embodiment can also not depart from present inventive concept within the knowledge of a person skilled in the art
Under the premise of make a variety of changes.
Claims (5)
1. a kind of optimization method based on Stentiford vision modes, which is characterized in that include the following steps：
Step 001. carries out gray proces for pending coloured image, obtains pending gray level image, and enter step 002；
Each pixel that step 002. is directed in pending gray level image respectively sets the pixel centered on pixel
Corresponding round measuring and calculating region, wherein the radius that the circle calculates region is the distance between neighbor pixel, which surveys
It includes the pixel and up and down four pixel adjacent with the pixel to calculate region；Thus to obtain pending gray-scale map
The corresponding round measuring and calculating region of each pixel difference as in, subsequently into step 003；
Step 003. is directed to each pixel in pending gray level image respectively, will be in round measuring and calculating region corresponding to pixel
Up and down four pixel adjacent with the pixel, and on the circumference that the circle calculates region, neighbor pixel
Between centre position obtained as eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel
The pixel value of eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel, subsequently into step 004；
Step 004. is directed to each pixel in pending gray level image respectively, for round measuring and calculating region corresponding to pixel
In eight objects adjacent with the pixel, whether the pixel value for being respectively compared each object be more than or equal to the pixel of the pixel
Value, it is then 1 for the object tag to be, it is 0 to be otherwise directed to the object tag；Then it is directed to the mark value of eight objects, is pressed
Default initial position and sequence, which are combined, constitutes an eight bit, and the eight bit is converted to the decimal system
Numerical value corresponds to the pixel；Thus obtain each pixel in pending gray level image respectively corresponding one ten into
Numerical value processed, enters step 005；
Step 005. is according to pending gray level image, using Uniform LBP patterns, by each pixel in pending gray level image
The corresponding decimal value of point difference, is converted to Uniform that is corresponding with pending gray level image, including 59 kinds of numerical value
LBP matrixes, wherein the number of element is equal with the number of pixel in pending gray level image in Uniform LBP matrixes, and
It corresponds, the value of each element is the grade point as corresponding pixel points, Uniform LBP matrixes in Uniform LBP matrixes
A total of 59 kinds of numerical value of value of middle element, i.e., a total of 59 kinds of the grade point of pixel in pending gray level image；Subsequently into
Step 006；
Uniform LBP matrix of the step 006. corresponding to pending gray level image, in pending gray level image
Each pixel establishes the three-dimensional array [Rank, Value, Location (x, y)] corresponding to pixel, wherein Rank respectively
Indicate that the grade point of corresponding pixel points, Value indicate that the pixel value of corresponding pixel points, Location (x, y) indicate respective pixel
The coordinate of point；Subsequently into step 007；
Three-dimensional array of the step 007. corresponding to each pixel in pending gray level image, will be in pending gray level image
All pixels press the grade point Rank corresponding to it and divided, carry out statistics with histogram, that is, obtain each grade point Rank
The corresponding pixel collection R of institute*, subsequently into step 008；
Step 008. is directed to 59 kinds of grade point Rank, by minimum level value Rank, by grade point Rank from as low as big suitable
Sequence is sequentially directed to the pixel collection R corresponding to each grade point Rank respectively*, for pixel collection R*In each pixel
Point 00801 is operated to step 00803, and then obtains each pixel in pending gray level image as follows respectively
Significance；
Step 00801. pixel collection R where pixel*In in other each pixels in addition to the pixel, arbitrarily
It takes out and presets N number of pixel, the set T corresponding to the pixel is formed, subsequently into step 00802；Wherein, N is less than or equal to
Each pixel collection R*In minimum pixel number subtract 1；
Step 00802. is directed to the pixel, count its correspond in set T with the unequal pixels of pixel pixel value Value
The number P of point, and enter step 00803；
Step 00803. calculatesObtain the significance of the pixel.
2. a kind of optimization method based on Stentiford vision modes according to claim 1, which is characterized in that the step
Further include step 009 after rapid 008, after executing the step 008, enters step 009, wherein step 009 is as follows：Step 009.
For the significance of each pixel in pending gray level image, it is multiplied by preset ratio coefficient respectively, updates pending gray-scale map
The significance of each pixel obtains corresponding specific image further according to pending gray level image as in.
3. a kind of optimization method based on Stentiford vision modes according to claim 2, it is characterised in that：It is described pre-
If proportionality coefficient is 255.
4. a kind of optimization method based on Stentiford vision modes according to claim 1, which is characterized in that the step
Rapid 003 specifically include it is as follows：
The each pixel being directed to respectively in pending gray level image, according to the picture of adjacent four pixels up and down of pixel
Element value, and the upper left adjacent with the pixel, upper right, lower-left, four pixels in bottom right pixel value, obtain the circle respectively
Calculate region circumference on, between neighbor pixel centre position pixel value；Then calculate round corresponding to the pixel
In region up and down four pixels adjacent with the pixel and the circle measuring and calculating region circumference on, adjacent pixel
Centre position between point, as eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel, i.e.,
Obtain the pixel value of eight objects adjacent with the pixel in round measuring and calculating region corresponding to the pixel.
5. a kind of optimization method based on Stentiford vision modes according to claim 4, which is characterized in that the step
In rapid 003, each pixel being directed to respectively in pending gray level image, according to adjacent four pixels up and down of pixel
Pixel value, and the upper left adjacent with the pixel, upper right, lower-left, four pixels in bottom right pixel value, it is double using four times
Linear interpolation method, obtain respectively the circle measuring and calculating region circumference on, between neighbor pixel centre position pixel value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510926402.9A CN105550690B (en) | 2015-12-14 | 2015-12-14 | A kind of optimization method based on Stentiford vision modes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510926402.9A CN105550690B (en) | 2015-12-14 | 2015-12-14 | A kind of optimization method based on Stentiford vision modes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105550690A CN105550690A (en) | 2016-05-04 |
| CN105550690B true CN105550690B (en) | 2018-09-11 |
Family
ID=55829873
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510926402.9A Active CN105550690B (en) | 2015-12-14 | 2015-12-14 | A kind of optimization method based on Stentiford vision modes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105550690B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093469A (en) * | 2013-01-22 | 2013-05-08 | 清华大学 | Depth extraction method based on visual model |
| CN103295206A (en) * | 2013-06-25 | 2013-09-11 | 安科智慧城市技术（中国）有限公司 | low-light-level image enhancement method and device based on Retinex |
| CN104915972A (en) * | 2014-03-13 | 2015-09-16 | 欧姆龙株式会社 | Image processing apparatus, image processing method and program |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007012798A1 (en) * | 2005-07-28 | 2007-02-01 | British Telecommunications Public Limited Company | Image analysis |
-
2015
- 2015-12-14 CN CN201510926402.9A patent/CN105550690B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093469A (en) * | 2013-01-22 | 2013-05-08 | 清华大学 | Depth extraction method based on visual model |
| CN103295206A (en) * | 2013-06-25 | 2013-09-11 | 安科智慧城市技术（中国）有限公司 | low-light-level image enhancement method and device based on Retinex |
| CN104915972A (en) * | 2014-03-13 | 2015-09-16 | 欧姆龙株式会社 | Image processing apparatus, image processing method and program |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105550690A (en) | 2016-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Saavedra | Sketch based image retrieval using a soft computation of the histogram of edge local orientations (s-helo) | |
| CN103207879B (en) | The generation method and apparatus of image index | |
| Srivastava et al. | A review: color feature extraction methods for content based image retrieval | |
| CN108898643B (en) | Image generation method, device and computer readable storage medium | |
| CN110413816B (en) | Color Sketch Image Search | |
| Sun et al. | Monte Carlo convex hull model for classification of traditional Chinese paintings | |
| Casanova et al. | Texture analysis using fractal descriptors estimated by the mutual interference of color channels | |
| CN113392856B (en) | Image forgery detection device and method | |
| US8970593B2 (en) | Visualization and representation of data clusters and relations | |
| CN104504723A (en) | Image registration method based on remarkable visual features | |
| CN102629325B (en) | Image characteristic extraction method, device thereof, image copy detection method and system thereof | |
| Mishra et al. | Medical image retrieval using self-organising map on texture features | |
| Lazarou et al. | A novel shape matching descriptor for real-time static hand gesture recognition | |
| CN114419265A (en) | Interactive map display method based on map image recognition | |
| CN105550690B (en) | A kind of optimization method based on Stentiford vision modes | |
| Liang et al. | Sketch-based retrieval using content-aware hashing | |
| Debroutelle et al. | Automatic classification of ceramic sherds with relief motifs | |
| Blanc-Beyne et al. | A salience measure for 3D shape decomposition and sub-parts classification | |
| Yang et al. | Color image representation using invariant exponent moments | |
| CN114581389A (en) | Point cloud quality analysis method based on three-dimensional edge similarity characteristics | |
| Yang et al. | Importance-based approach for rough drawings | |
| He et al. | A computational fresco sketch generation framework | |
| Amiri et al. | Copy-move forgery detection using a bat algorithm with mutation | |
| CN106570125B (en) | Remote sensing image retrieval method and device for rotational scaling and translation invariance | |
| Liu et al. | New anti-blur and illumination-robust combined invariant for stereo vision in human belly reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20191225 Address after: Room 108, floor 1, No.28, ningshuang Road, Yuhuatai District, Nanjing City, Jiangsu Province Patentee after: Nanjing fortune Health Industry Co., Ltd. Address before: Yuen Road Qixia District of Nanjing City, Jiangsu Province, No. 9 210023 Patentee before: Nanjing Post & Telecommunication Univ. |