CN109035317A - Illumination reversion and invariable rotary texture expression based on three value mode of gradient local - Google Patents

Illumination reversion and invariable rotary texture expression based on three value mode of gradient local Download PDF

Info

Publication number
CN109035317A
CN109035317A CN201810724136.5A CN201810724136A CN109035317A CN 109035317 A CN109035317 A CN 109035317A CN 201810724136 A CN201810724136 A CN 201810724136A CN 109035317 A CN109035317 A CN 109035317A
Authority
CN
China
Prior art keywords
value
reversion
gradient
lbp
illumination
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.)
Pending
Application number
CN201810724136.5A
Other languages
Chinese (zh)
Inventor
宋铁成
辛亮亮
张刚
罗忠涛
张天骐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201810724136.5A priority Critical patent/CN109035317A/en
Publication of CN109035317A publication Critical patent/CN109035317A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G06T7/41—Analysis of texture based on statistical description of texture
    • 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

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to a kind of illumination reversion and invariable rotary texture expression based on three value mode of gradient local, the following steps are included: first, a width gray level image is inputted, the gradient value of center pixel and neighbor pixel is calculated, adaptive threshold is introduced and three value quantizations is carried out to gradient value；Then, three value modes after quantization are divided into positive binary pattern and negative binary pattern, calculate separately positive LBP coding and negative LBP coding with rotational invariance；Finally, calculating the statistic histogram of positive LBP coding and negative LBP coding, and cascaded as final feature vector.The method of the invention has good robustness to illumination reversion and the rotationally-varying of image, and texture ability to express of traditional LBP method under illumination reversion variation can be improved.

Description

Illumination reversion and invariable rotary texture expression based on three value mode of gradient local
Technical field
The present invention relates to Digital Image Processing, computer vision field, and in particular to one kind is based on three value mould of gradient local The illumination reversion and invariable rotary texture expression of formula.
Background technique
The textural characteristics of image reflect the space distribution rule of pixel grey scale, conveyed the surface structure information of object. The underlying issue that effective textural characteristics are Digital Image Processing and computer vision field research is extracted, in Texture classification, line It plays an important role in the visual tasks such as reason segmentation, scene Recognition, images match.Therefore, texture characteristic extracting method has weight The researching value wanted, scholars propose many texture characteristic extracting methods.Wherein representative method has: gray level co-occurrence matrixes (Gray-Level Co-occurrence Matrix, GLCM), Markov random field (Markov Random Field, MRF), method and the method based on local mode popular in recent years based on primitive (texton) study.
Local binary patterns (Local Binary Pattern, the LBP) method of the propositions such as Ojala is by encoding local picture The symbolic information of plain difference establishes histogram spy, have computation complexity is low, linear gradation invariance, without training study and The neck such as it is easy to many merits such as Project Realization, and is widely used in Texture classification, recognition of face, image retrieval and pedestrian detection Domain.Thought based on LBP, scholars propose many derivative algorithms, including three value mode (Local Ternary of part Pattern, LTP), complete LBP (Complete LBP, CLBP), leading LBP (Dominant, DLBP) etc..
Although the method based on LBP thought has many advantages, such as, tradition LBP and its deriving method invert illumination and change It is very sensitive.Illumination reversion variation is due to reflection, the complicated illumination condition such as exposes, blocks certain parts of image is caused to become It is bright, and some Dimming parts.The relative size that illumination reversion variation will cause gray value of image changes, so as to cause part The symbol of pixel difference changes.Once illumination reversion variation occurs, this can seriously affect LBP and its derivative algorithm and classify With the performance in tasks such as identification.Against the above deficiency, the invention proposes one kind to be based on three value mode of gradient local The illumination reversion and invariable rotary texture expression of (Gradient Local Ternary Pattern, GLTP).
Summary of the invention
The illumination based on three value mode of gradient local that the problem to be solved by the invention is to provide a kind of is inverted and is rotated not Become texture expression, the method has good robustness to illumination reversion and the rotationally-varying of image, tradition can be improved Texture ability to express of the LBP method under illumination reversion variation.
It is as follows that the technical solution to solve the above problems is invented herein: a kind of illumination reversion based on three value mode of gradient local With invariable rotary texture expression, comprising the following steps:
Step 1, a width gray level image is inputted, its gray value is normalized into [0,255]；
Step 2, neighbor pixel samples: utilizing the neighbor pixel of each center pixel of interpolation calculation；
Step 3, the gradient value of each center pixel Yu its neighbor pixel is calculated, introduces adaptive quantizing threshold value to gradient value Carry out three value quantizations；
Step 4, three value modes after quantization are divided into positive binary pattern and negative binary pattern, and calculate separately to have and rotates The positive LBP of invariance is encoded and negative LBP coding；
Step 5, the statistic histogram of positive LBP coding and negative LBP coding is calculated separately, and two histograms are cascaded into conduct Final feature vector.
Invention has the advantage that compared with prior art herein
First, the present invention is based on the thought of traditional LBP, the gradient informations of encoding centre pixel and neighbor pixel, to illumination Reversion variation has good robustness；
Second, compared with traditional LBP, the gradient information of center pixel and neighbor pixel is quantified as three value moulds by the present invention Formula increases algorithm to the robustness of noise；
Third, compared with the fixed quantisation threshold value of LTP setting, present invention introduces adaptive quantization thresholds, to linear light There is robustness according to variation.
Detailed description of the invention
Fig. 1 is that the present invention is based on the illumination reversions of three value mode of gradient local and invariable rotary texture to express method flow Figure.
Fig. 2 is that gradient value of the present invention calculates and three values quantization schematic diagram.
Fig. 3 is that the positive LBP coding of the present invention and negative LBP coding calculate schematic diagram.
Specific embodiment
The principle of the present invention is described further below in conjunction with attached drawing and specific implementation method.
Referring to Fig.1, the present invention is based on the illumination reversion of three value mode of gradient local and invariable rotary texture expression, packets Include following steps: input picture, neighbor pixel sampling, gradient value are calculated to be counted with the quantization of three values, positive LBP coding and negative LBP coding Calculation, counting statistics histogram simultaneously cascade.
Step 1, its gray value is normalized to [0,255] by input gray level image I.
Step 2, if the sample radius of circle shaped neighborhood region is r, the coordinate of center pixel is (x, y), gray value xc, can acquire P neighbor pixel x of center pixeli(i=0,1 ... P-1), the coordinate of neighbor pixel is calculated as (x+rcos (2 π i/P), y- rsin(2πi/P)).For not exclusively falling in the pixel on image lattice, gray value is estimated by bilinear interpolation.
Step 3, referring to Fig. 2, center pixel x at position (x, y) is definedcWith each neighbor pixel xiGradient value are as follows:
gi(x, y)=| xi-xc|, i=0,1 ..., P-1 (1)
After calculating the gradient value of center pixel and neighbor pixel, introduces adaptive threshold and three values are carried out to gradient value Quantization, quantization function may be expressed as:
In formula, τ is a parameter of experiment setting, and Δ is the average value of entire image gradient
In formula, M, N respectively indicate the length and width of image, from formula (2) as can be seen that quantization function sτQuantization threshold Change with the value of Δ, and is no longer a fixed value.Result after provable quantization is to linear light according to reversion changing pattern Type I '=aI+b (a < 0) has robustness.Have known to (1) and (2):
As can be seen that factor b is eliminated by the gradient operation in (1) formula, factor a is by (2) formula quantization function sτIt eliminates, because This, linear light is remained unchanged according to the quantized result of reversion variation front and back.
Step 4, after the quantized value for obtaining inverting illumination variation robust, it should encode, obtain to P quantized value It is encoded to GLTP
But higher (the dimension 3 of dimension of GLTP descriptor calculated in this wayP), in order to obtain rotational invariance and guarantor Lower characteristic dimension is held, three value modes after quantization are now divided into positive binary pattern and negative binary pattern, and calculate separately tool There are the positive LBP coding and negative LBP coding of rotational invariance (referring to Fig. 3)
(6) formula is positive LBP coding, and (7) formula is negative LBP coding, and subscript " ri " represents rotational invariance, and " u2 " indicates uniform Estimate U≤2, is respectively defined as
Sign function suAnd slEffect be the quantization of (1) gradient value calculated is positive binary pattern and negative binary pattern, It is respectively defined as:
As can be seen that compared to GLTP coding is directly calculated, after being divided into positive LBP coding and negative LBP coding, the feature of algorithm Dimension only has 2 (P+2), but also has rotational invariance.
Step 5, the statistic histogram of positive LBP coding and negative LBP coding is calculated separately, and two histograms are cascaded into conduct Final feature vector.

Claims (3)

1. a kind of illumination reversion and invariable rotary texture expression based on three value mode of gradient local, which is characterized in that packet Include following steps:
Step 1, a width gray level image is inputted, its gray value is normalized into [0,255]；
Step 2, neighbor pixel samples, and utilizes the neighbor pixel of each center pixel of interpolation calculation；
Step 3, the gradient value of each center pixel Yu its neighbor pixel is calculated, adaptive quantizing threshold value is introduced and gradient value is carried out The quantization of three values；
Step 4, three value modes after quantization are divided into positive binary pattern and negative binary pattern, and calculated separately with invariable rotary Property positive LBP coding and negative LBP coding；
Step 5, the statistic histogram of positive LBP coding and negative LBP coding is calculated separately, and by two histogram cascades as final Feature vector.
2. method according to claim 1, which is characterized in that in the step 3, calculate center pixel xcWith each neighbour's picture The gradient value g of plain xii, and introduce adaptive threshold and three value quantizations are carried out to gradient value.
3. method according to claim 2, which is characterized in that in the step 3, quantization function sτQuantization threshold with Δ Value and change, and be no longer a fixed value；Result after provable quantization to linear light according to reversion variation model I '= AI+b (a < 0) has robustness: sτ(g′i, Δ ') and=sτ(|a|gi, | a | Δ)=sτ(gi,Δ)。
CN201810724136.5A 2018-07-04 2018-07-04 Illumination reversion and invariable rotary texture expression based on three value mode of gradient local Pending CN109035317A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810724136.5A CN109035317A (en) 2018-07-04 2018-07-04 Illumination reversion and invariable rotary texture expression based on three value mode of gradient local

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810724136.5A CN109035317A (en) 2018-07-04 2018-07-04 Illumination reversion and invariable rotary texture expression based on three value mode of gradient local

Publications (1)

Publication Number Publication Date
CN109035317A true CN109035317A (en) 2018-12-18

Family

ID=65521698

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810724136.5A Pending CN109035317A (en) 2018-07-04 2018-07-04 Illumination reversion and invariable rotary texture expression based on three value mode of gradient local

Country Status (1)

Country Link
CN (1) CN109035317A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110489587A (en) * 2019-07-31 2019-11-22 西安邮电大学 The tire trace image characteristic extracting method of three value mode of Local gradient direction
CN112819872A (en) * 2019-11-15 2021-05-18 瑞昱半导体股份有限公司 Image processing method based on sensor characteristics

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103778412A (en) * 2014-01-16 2014-05-07 重庆邮电大学 Face recognition method based on local ternary pattern adaptive threshold
CN105787492A (en) * 2016-04-01 2016-07-20 电子科技大学 Local ternary pattern texture feature extraction method based on mean sampling
CN107463917A (en) * 2017-08-16 2017-12-12 重庆邮电大学 A kind of face feature extraction method merged based on improved LTP with the two-way PCA of two dimension

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103778412A (en) * 2014-01-16 2014-05-07 重庆邮电大学 Face recognition method based on local ternary pattern adaptive threshold
CN105787492A (en) * 2016-04-01 2016-07-20 电子科技大学 Local ternary pattern texture feature extraction method based on mean sampling
CN107463917A (en) * 2017-08-16 2017-12-12 重庆邮电大学 A kind of face feature extraction method merged based on improved LTP with the two-way PCA of two dimension

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ROSS P. HOLDER AND JULES R. TAPAMO: "Improved gradient local ternary patterns for facial expression recognition", 《EURASIP JOURNAL ON IMAGE AND VIDEO PROCESSING》 *
TIECHENG SONG ET AL.: "Grayscale-Inversion and Rotation Invariant Texture Description Using Sorted Local Gradient Pattern", 《IEEE SIGNAL PROCESSING LETTERS》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110489587A (en) * 2019-07-31 2019-11-22 西安邮电大学 The tire trace image characteristic extracting method of three value mode of Local gradient direction
CN110489587B (en) * 2019-07-31 2023-04-28 西安邮电大学 Tire trace image feature extraction method in local gradient direction three-value mode
CN112819872A (en) * 2019-11-15 2021-05-18 瑞昱半导体股份有限公司 Image processing method based on sensor characteristics
CN112819872B (en) * 2019-11-15 2024-05-24 瑞昱半导体股份有限公司 Image processing method based on sensor characteristics

Similar Documents

Publication Publication Date Title
Wang et al. Single image dehazing based on the physical model and MSRCR algorithm
Xiaofeng et al. Discriminatively trained sparse code gradients for contour detection
Long et al. Identifying natural images and computer generated graphics based on binary similarity measures of PRNU
CN111079764B (en) Low-illumination license plate image recognition method and device based on deep learning
CN104008364B (en) Face identification method
CN112270317B (en) Reading identification method of traditional digital water meter based on deep learning and frame difference method
CN110211127B (en) Image partition method based on bicoherence network
CN112614110B (en) Method and device for evaluating image quality and terminal equipment
CN113139544A (en) Saliency target detection method based on multi-scale feature dynamic fusion
CN102695056A (en) Method for extracting compressed video key frames
CN109743642A (en) Video abstraction generating method based on layered circulation neural network
CN109035317A (en) Illumination reversion and invariable rotary texture expression based on three value mode of gradient local
CN116030396B (en) Accurate segmentation method for video structured extraction
Tao et al. Smoke vehicle detection based on spatiotemporal bag-of-features and professional convolutional neural network
Tan et al. Detecting computer generated images based on local ternary count
CN101510265B (en) Association moving number character intelligent identification method
CN111627047B (en) Underwater fish dynamic visual sequence moving target detection method
CN102930558A (en) Real-time tracking method for infrared image target with multi-feature fusion
Ma et al. Background subtraction based on multi-channel SILTP
Saifullah et al. Fish detection using morphological approach based-on K-means segmentation
CN117115600A (en) No-reference image quality evaluation method and device and electronic equipment
CN103871084B (en) Indigo printing fabric pattern recognition method
CN114926348A (en) Device and method for removing low-illumination video noise
CN108109141B (en) Homologous local replication detection method based on superpixel multi-feature matching
WO2020124390A1 (en) Face attribute recognition method and electronic device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20181218