TW201227534A - Method and system for matching texture feature points in images - Google Patents

Abstract

A method and a system for matching at least two texture feature points in at least two images are provided. The method for matching the texture feature points includes the following steps. A reference image and a target image are received. A LBP reference image is obtained according the reference image and a LBP target image is obtained according the target image. A plurality of reference texture feature points in the LBP reference image are detected. A plurality of target texture feature points in the LBP target image are matched according the reference texture feature points.

Description

201227534 1 W&ybbFiOi VI. Description of the Invention: [Technical Field of the Invention] The present invention relates to an image comparison method and system, and more particularly to a texture feature point comparison method and system for an image. [Prior Art] Nowadays, the more common techniques for feature point extraction are corner detection. There are many methods for detecting corner points. The purpose is to find the location of feature points with more discrimination in grayscale images, hoping to eliminate easy comparison. The wrong line and the brightness value change are more consistent areas, and the extracted feature points are usually the areas where the gray image brightness value is stronger and more corners. Although the effect is stable, the number of feature points detected is Not dense enough, in fact, there are still many regions in the grayscale image, the degree of discrimination of the comparison is also quite high. The traditional method of comparison has the optical flow method and a large number of block comparison methods. The traditional optical flow method is limited by the definition of its own theory, and cannot compare the feature points with larger movement amount, and has no inhibition ability for light changes. The speed is also slightly Φ too slow. SUMMARY OF THE INVENTION The present invention relates to an image comparison method and system, which utilizes a block comparison method, which has a certain ability to suppress light changes. According to the first aspect of the present invention, a texture feature point comparison method for an image is proposed. The texture feature point comparison method includes the following steps. Receive a reference image and a target image. According to the reference image, a local Binary Pattern (LBP) reference image is generated and according to the target 201227534

The TW6966PA image generation - LBP target image. Measure the LBp reference, texture feature points. According to the reference texture feature points, the corresponding target texture feature points are compared. According to the second aspect of the present invention, a comparison system is proposed. The texture feature point comparison system includes a Loca Binary mtern (cafe) generating a unit shape f-aligning unit. Provide unit (4) pure - mention = early reference image and - target image. £βρ generation unit 2 = - participation - LBP reference image, and according to the target reference shirt image, the image looks for the corresponding number of target texture feature points Feature point 'in the target image' and cooperate with: Figure::: More: solution, the following special implementation [implementation] Please refer to Figure 1 and Figure 2, texture feature point comparison method The flow chart of the texture feature point comparison system of the first embodiment of the present invention is 〇〇==Fig. Green. The image point comparison method of the embodiment mainly includes three parts and eight figures. The texture of this embodiment Special extraction, (1) partial binary graphics (L〇c^. (1) block comparison of tatographic features of texture feature points and (3) lnary Pattern, LBP), Difference, SAD) 抡, =, difference and Sum of Absolute image texture point matching system with the shadow of the second picture belongs to the technical field The average person who has the usual knowledge:: ratio = . Forgot. 4 201227534 The method is not limited to the comparison system 100 of Figure 2, nor is it limited to the following calculation examples. (1) Texture feature points As shown in FIG. 1 , in step S101 , the providing unit 110 provides a reference image Ir and a target image It for receiving by the texture feature point comparison system 100 of the image of the embodiment. The providing unit 110 is, for example, a camera, a camera or a storage device for storing a plurality of images. The reference image Ir and the target image It are, for example, a moment before the continuous shooting and the current φ image. Next, in step S102, the LBP generating unit 120 is based on the reference image Ir. An LBP reference image LBPr is generated, and an LBP target image LBPt is generated according to the target image It. For example, the LBP reference image LBPr of the embodiment is generated by performing a 3x3 mask operation on the reference image Ir. As a result, the 3x3 mask operation method compares the brightness of the reference pixel Pr of the center point of the mask with the brightness of the surrounding eight reference pixels Pr. • Refer to FIG. 3, which shows L. BP operation diagram. Reference image

The luminance of one reference pixel Pr of Ir can be expressed by 8 bits, which are denoted as bl, b2, b3, b4, b5, b6, b7, b8, respectively. When the brightness of the adjacent reference image Pr is greater than the brightness of the center reference pixel Pr, the value is set to 1; otherwise, the value is set to 0. As shown on the right side of Figure 3, the last 8-bit LBP value is "11111100". The LBP reference image LBPr is obtained after each of the reference pixels Pr of the reference image Ir calculates its LBP value. Referring to FIG. 1, a plurality of original images and their LBP images are illustrated. 201227534

TW6966PA The upper row of images in Figure 1 is not brightness."

The image is its corresponding LBP image. From the attached image: Figure 1 below, a kind of regional brightness contrast relationship, it can be known that the LBP image has a considerable degree of tolerance. When $ or darker, the light change image does not change much when the light of the starter image changes, and the LBP then 'in step S103, prepares the remaining _ like Ir's several target textures. Before extracting the reference shadow method, first explain the debt measurement unit ^ to describe the texture feature point extraction (H_ - Stance) to obtain two Glb=== month distance in the LBP image, a difference in value. ::r size, _===; 2 lines of binary octets and: Otherwise, the result of all, == ° will have a 〇 result, the value is 丨 _ number, repulsion ((4) logic operation is as follows (1) For example, "10110111" and Γ1011ιηιη XOR 10111010 οοοοΰοι · : = The result is that the Hamming distance is 3. = The greater the distance from the U Hamming, the greater the difference in the table (four) ► · · · · Lu ♦ ... ..........................C1) Please refer to Figure 4, its Ranganquanquan reference pixel as a 6): meaning: test: Reference test f 7X7 reference like reading _ 'Cai Kong Pixel Pro Shooting: a circle C, and there will be 16 reference pixels xi around this circle c = 201227534

I W6966PA (i=l~16). Detection unit 13. And set a side: value Mo: from ^ Hamming distance U foot _ y, shame 2 although continuous η: a reference texture feature point we want to practice. The test is as follows: When the vicinity of the pixel Pr 〇 and the reference image and the reference pixel Pr0 are smooth regions, the H is large. So when it is lower. And Hamming distance u foot " two 'mu indicates that the continuous number of geometric angles of this reference pixel Pr0 can be Pr 〇 corner point, its Hamming distance (four) is full; " column such as 'pixel simple straight edge, It: the number of consecutive _%> 丨 连续 will be relatively small. In the ~ 禹 130 t: parameter ° to rule out the easy comparison of the wrong shirt side: the measurement unit clearly refers to Figure 2', which shows the reference texture feature point edge of the above-mentioned article ^, can be _ for the reference texture foot ( 2) Block alignment of the LBP image. Then, in step S104, the 罝*1/ln texture feature point fr is compared to the target, such as If t 140, according to the reference feature points ft. , ι out the corresponding number of target lines to find ^ LB! ^ 生早疋 12G get LBP reference image coffee tray LBP 2 2 test list f 13Q _ out reference texture feature point target texture feature f"t仃可== Set the comparison ^ the target image It's point ^ corresponds to the target texture feature point ft ^ out reference texture feature 201227534

TW6966PA * j For example, please refer to Figure 5, which shows the LBP block alignment. The matching unit 140 finds a position (^7) corresponding to a reference texture feature point fr at the target image It. Next, with the position as the center, the 7χ7 search range R is selected in the target image It box. The comparison unit 14 〇 and in the reference image Ι γ, selects the 3 χ 3 reference alignment block rl with the reference texture feature point fr as the center. The comparison unit 140 is further in the 7x7 search range R, and the target 3x3 target comparison block r2 is selected for each target pixel pt. Thus, 7 π target contrast blocks r2 can be selected (the fifth graph only shows one target comparison block r2). Then, the comparison unit 140 calculates the reference comparison block rl and each target alignment area. The sum of the Hamming distances of the block r2. Its mathematical expression is expressed by the following formula (2) · 廿... ............(2) Eight middle Μ is the moving vector '4U', „e[_33], ve[Sister Seven” force is The reference image Ir towel, the coordinate value of the reference pixel & (x + i, w). The taste _, _ν is the LBP value of the reference pixel Pt whose coordinates are + a) in the target image It. When counting all the reference pixels in the 7x7 search range and the Hamming distance κ„, ν) of the target Pt, set the _ gate value. When the Hamming ^ is called less than this threshold, 'represent this reference. The similarity between the pixel 阡 and the prime Pt is high enough, then the target image is set to be unrecognized... the candidate point, and finally... the search range 201227534 woyoom point) the motion vector of the texture feature point and the motion vector of the plurality of possible target texture feature points ft ( The set of wv). The bamboo block (? Absolute error sum (SAD) block comparison: texture: ί point step is: medium, the judgment unit 150 judges that each reference only corresponds to one target texture feature point ft. If fi, Then: the texture feature point ^ corresponds to a plurality of target texture feature points it, then proceeds to step Si 〇 6. Also fr 160 (' 〇uUu e D Γ ' 'point ^ the absolute error of the brightness and the standard texture feature point ft (four) (10) The heart selects one of the items un - for example, 'after the above LBP 11 block alignment, the comparison single-= to the coordinates is (the texture feature point fr may be opposite the shore t = =)' corresponds to multiple target textures Feature point two rationality = two: solid reference texture feature point kg corresponds to - a target bamboo (four) u u卩Real motion vector (mv). 16〇(SAD) is a global search method, which selects the unit into the reference texture feature point and its corresponding minimum axis vector. (―).- The objective of the feature point (10), whose coordinates are (^W*,^ + V*)=arg SAD{l>y)(x + U}y + v)....... ................... 运运=====Off: In fact, 'I will use the way of looking up the table directly to calculate j 201227534

The number of TW6966PA. As shown in Fig. 2, the storage unit 170 can establish an 8-bit Hamming distance table whose data table size is 8x256 = 2, 048 bits, as shown in Table 1 below: Hamming distance of the XOR operation is the first 2nd 3rd, 4th, 5th, 6th, 8th, 8th, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th, 1st, 8th 1 1 0 0 0 0 0 0 1 0 1 0 0 0 0 0 —0 1 1 2

-----^~L_ί_____ The table lookup unit 180 can calculate the 32-bit XOR operation according to the Hamming distance table to find the Hamming distance, and only needs 4 times to find the name to complete quickly, which contributes to the overall performance. Improvement. In this embodiment, the block comparison method is used, because the use of ^   experiment (4) 'has a certain suppression force for light changes, and the data structure of the feature image in the feature point extraction method on the beach is matched with In the way of 'significantly reduce the calculation of the seam compared with the general block comparison method: 201227534 1 wovoor / x has a relatively good performance. In summary, although the present invention has been disclosed above by way of example, it is not intended to limit the present invention. Those who have ordinary knowledge in the technical field of the present invention can make various changes and refinements without departing from the spirit and scope of the case. Therefore, the scope of protection in this case is subject to the definition of the scope of the patent application attached. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a flow chart showing a method for comparing texture feature points of an image of the present embodiment. Fig. 2 is a block diagram showing the texture feature point comparison system of the image of the embodiment. Figure 3 shows a schematic diagram of the LBP operation. Figure 4 is a schematic diagram of a reference pixel and surrounding 16 reference pixels. Figure 5 is a schematic diagram showing the comparison of LBP blocks. φ Figure 1 shows several original images and their LBP images. 2 is a schematic diagram of a reference texture feature point. [Main Component Symbol Description] 100: Texture Feature Point Alignment System 110: Providing Unit 120: LBP Generating Unit 130: Detecting Unit 140: Aligning Unit 201227534

TW6966PA 150 : Judging unit 160 : Selection unit 170 : Storage unit 180 : Table lookup unit bit b b b ' b3, b4 ' b5 ' b6 ' b7, b8 C : Circle fr : Reference texture feature point ft, ft * : target Texture feature point Ir: Reference image It: Target image LBPr: LBP reference image LBPt: LBP target image Pr, PrO, xi: Reference pixel Pt: Target pixel R: Search range r 1 : Reference comparison block r2: Target comparison Blocks S101 to S106: Flow Step 12

Claims (1)

201227534 i we>y (10) wv· VII. Patent application scope: ^-An image texture feature point comparison method, including: receiving a reference image and a target image; according to the reference image, producing 'Binary Pattern, LBP) reference $ stupid.邛 兀 兀 ( ( ( ( ( ( ( L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L Texture feature point....Immediately compares 2. As in the patent application scope comparison method, in the step of generating the texture feature point of the image of the LBp parameter, > the shirt image and the target of the LBP target image The image includes the relationship of the number of swims and the size of the pixels of the test pixel; the image of the LBP target image for each target phase and the adjacent ones. 3. For the patent range, the brightness relationship of the primes. The LBp refers to the texture feature points of the image of H, and detects the reference texture features ^ the number of LBP reference images * ^ ^ -LBP points.) Detecting the reference texture features as described in the item The Hamming distance of the image feature point pixels of the image is checked by the == near the LBP reference 13 V 201227534 1 W6966FA fl 5. The texture feature point comparison method of the image according to claim 1 of the patent scope LBP The reference image includes a plurality of LBP reference pixels. The LBP target image includes a plurality of LBP target pixels, and the step of comparing the corresponding binocular texture feature points is based on the reference pixels and the LBM target pixels. The Hamming distance is compared to the corresponding target texture feature points. 6. The texture feature point comparison method of the image of claim 5, wherein the Hamming distances of the LBp reference pixels and the LBp target pixels are obtained by using a look-up table. 7. The method for comparing texture feature points of the image according to the third aspect of the patent scope includes: determining whether each reference texture feature point corresponds to only one target texture feature point; and one of the right reference texture feature points Corresponding to the plurality of target texture feature points, the towel-target texture feature points are selected according to the brightness of the reference texture feature points and the absolute error sum (Sum 〇f Absolute Difference, SAD) of the target texture feature points. A 08. A texture feature point comparison system for receiving an image-reference image and a target image provided by a providing unit, comprising: a local Binary Pattern (LBP): the second system is based on the reference The image generates an LBP reference image, and Yikang generates an LBP target image by using the private image; the texture single is a plurality of reference ten units of the reference image, according to the reference texture feature points, the L is 201227534 w〇y〇 The or/i target should be the texture tough point. Alignment system, ^ special - (d) 8 items _ reading feature point b reference shirt image includes a plurality of references for each reference pixel and the adjacent two 2 = image system the target cont "m pixel party size relationship For each target two; = target pixel 'The LBP target image system ίο如二, the brightness relationship of the target pixel. Point comparison system, the texture characterization of the image described in the second and second L items, the detection unit Weiyi The X 摅 ^ shirt image includes a plurality of LBP reference image reference materials, the BP reference pixel and the adjacent LBP points are compared with the texture features of the image described in item (7) of the ST range for storing - Hamming distance And a Hamming distance of the LBP reference pixels adjacent to the LBP reference image according to the Hamming distance table. The reference point is compared with the texture of the image described in item 8 The eigen element, the current LBP coffee reference image includes a plurality of LBP reference images ==:::, the target pixel, the comparison: the point. 卩 the corresponding target texture feature points ratio: Department: Shen = range 12th item The texture feature of the image 15 201227534 TW6966PA Unit for storing. 'A look-up table unit, based on the talk:, the monthly distance table; and the test pixel and the LBP target image = distance table to check out the LBP parameters 14. If the patent application / the = distance The point comparison system further includes: a texture feature of the image, a judging unit for judging a target texture feature point; and whether the reference texture feature point is only for a selected unit, if two of the target textures For the feature points, the :::::: corresponds to the brightness and the absolute error of the brightness of the feature points of the texture feature points of the reference texture (sum of Abs) Among them - target texture special = point 0