JP2012127871A - Texture quantification device and texture quantification method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quantify texture by joint transform correlation operation (JTC).SOLUTION: A texture quantification device 1 includes: a laser diode 2 for outputting laser light; an objective lens 3; a pin hole part 4 having laser beam as a point light source; a collimator lens 5 for providing parallel light beam with even wave-front phases; a polarizer 6; a half mirror 7; an optical computing correlation part 12 which has a spatial optical modulator 8 for inputting an image where a search histogram and a reference histogram are juxtaposed, performing spatial optical modulation on the basis of the image and generating an optical image, a Fourier transformation lens 10 for receiving the optical image from the spatial optical modulator 8 via the half mirror 7 and a polarizer 9, performing optical Fourier transformation to it and forming a Fourier transformation optical image, and a CMOS camera 11 for detecting the Fourier transformation optical image from the Fourier transformation lens 10 as an image; and a texture quantification part 13 for collating the search histogram and the reference histogram on the basis of intensity of a mutual correlation spot to determine similarity of the texture.

Description

  The present invention relates to a texture quantification apparatus and a texture quantification method for quantifying the texture of an object surface by paying attention to the roughness and gloss of the object surface.

  Various apparatuses for measuring the roughness of an object surface have been proposed. For example, a surface measurement method using a Fourier transform using a probe (Patent Document 1), a surface state display device for steel using a light receiving device (Patent Document 2), and a surface shape measuring device such as a magnetic tape using a Fourier transform using a sensor (Patent Document 1) Reference 3), strip-shaped steel plate surface inspection device using a histogram that receives reflected light (Patent Document 4), coating layer surface inspection device using a camera using a histogram (Patent Document 5), surface roughness analysis device for a magnetic recording medium (Patent Document 5) Patent Document 6).

Japanese Patent Laid-Open No. 51-149052 JP 52-49856 A JP 59-168305 A JP 59-180346 A Japanese Patent Laid-Open No. 62-12806 JP-A-9-79839

  However, the conventional invention simply measures the roughness of the surface of the object, and it is impossible to quantify the texture such as gloss, making it difficult to analyze a multifaceted surface.

  The present invention has been made to solve the above-described problems, and an object thereof is to quantify the texture of an object surface such as roughness and gloss.

  With the recent development of the robot market, visual process systems for robots similar to humans have been studied. Human visual recognition consists of real-time processing and high-precision recognition. The real-time processing recognizes the type of object by comparing visually recognized data with reference data. In conventional methods, it takes time to visually identify an object, so a high-speed image identification method is required to perform real-time processing like a person. As a high-speed image identification method, an optical correlator that can identify an image at high speed by image matching has attracted much attention. Image identification using an optical correlator can be performed at higher speed and faster than computer.

  Many studies of image identification using optical correlators have focused on shape identification used for fingerprint authentication systems, face authentication systems, and military tracking systems. However, there is a problem that the identification accuracy is low, and in order to recognize with high accuracy like a person, it is necessary to consider other elements in addition to the shape. The image has a lot of information such as color and texture in addition to the shape, and such information is important for the visual processing system. High-precision recognition is possible by shape, color, texture, and the like. The present invention is an identification technology invention relating to the texture. Here, the texture is caused by reflection, transmission, and diffusion, which are optical actions on the surface of the object. It recognizes the gloss of an object by reflection (for example, diamond), the transparency of the object by transmission (for example, glass), the rough feeling by diffusion, the softness (file, cloth), etc. with high accuracy.

  Therefore, the texture quantification device of the present invention uses the joint transformation correlation method (JTC), and the surface image captured by an imaging device such as a digital camera is converted into a luminance histogram. Various textures, such as smooth or rough surfaces, displayed on the modulator are quantified using the shape of the luminance histogram and a similar reference luminance histogram is selected. From these results, the quantification of the texture was realized using JTC.

  That is, the invention according to claim 1 is a spatial light modulator that modulates light from a light source, a Fourier transform lens that receives the modulated light from the spatial light modulator and performs Fourier transform, and light from the Fourier transform lens. A camera that captures an image, and performs a combined conversion correlation calculation of an input image input to the spatial light modulator to generate an output image, and the input image is set as a luminance histogram. A search histogram to be searched and a reference histogram that is set as a luminance histogram and created in advance for collation are set side by side, the input image is output to the spatial light modulator, and the combined transformation correlation calculation An output image including the autocorrelation spot and the cross-correlation spot obtained from the camera is obtained based on the intensity of the cross-correlation spot The collating the reference histogram and the search histogram, characterized in that and a texture quantifying section for determining the similarity of textures.

  The invention according to claim 2 generates a first modulated light by the spatial light modulator to which the input image is input, and generates a first optical image by performing a Fourier transform of the first modulated light by the Fourier transform lens. The first light image is picked up by the camera as a first converted image, the first converted image is output to the spatial light modulator, and the second modulation modulated by the first converted image by the spatial light modulator. Generating light, performing a Fourier transform of the second modulated light by the Fourier transform lens to generate a second light image, and capturing the second light image with the camera as the output image, To do.

  According to a third aspect of the present invention, the optical correlation calculation unit receives the first modulated light from the first spatial light modulator and the first spatial light modulator, performs Fourier transform, and generates a first optical image. A first Fourier transform lens, a first camera that captures a first light image from the first Fourier transform lens and generates a primary conversion image, and a primary conversion image from the first camera is input to generate second modulated light. A second spatial light modulator, a second Fourier transform lens that receives the second modulated light from the second spatial light modulator and performs a Fourier transform to generate a second light image, and a second Fourier transform lens from the second Fourier transform lens. And a second camera that captures a second light image and generates the output image.

According to a fourth aspect of the present invention, an input image in which a search histogram that is set as a luminance histogram and is an object of quantification of a texture and a reference histogram that is set as a luminance histogram and created in advance is placed in a spatial light modulator. A first correlation calculation step of inputting, Fourier-transforming the light from the spatial light modulator by a Fourier transform lens to form a first light image, capturing the first light image, and generating a primary conversion image; A primary conversion image captured in one correlation calculation step is input to a spatial light modulator, light from the spatial light modulator is Fourier transformed by a Fourier transform lens to form a second light image, and the second light image is converted into a second light image. A second correlation calculation step of imaging and producing an output image including an autocorrelation spot and a cross-correlation spot; and the search hysteresis based on the optical intensity of the cross-correlation spot of the output image. Compares the reference histogram gram, characterized by comprising the texture quantification determining the similarity of textures, the.
In the specific embodiment of the texture quantification apparatus and the texture quantification method described above, an image is obtained by imaging the surface layer of an object to be quantified, and is histogrammed with respect to luminance to create a search histogram. On the other hand, a reference histogram is set in advance as a database for collation. Then, the similarity of the texture is determined from the intensity of the cross-correlation spot obtained by performing the optical correlation calculation by the combined conversion correlation calculator using the search image in which the search histogram and the reference histogram are juxtaposed. According to such a configuration, it is possible to execute a texture search focusing on the texture with high accuracy and at high speed.

  For the luminance histogram indicating the texture, specifically, a texture distribution on a texture map in which the horizontal axis represents luminance and the vertical axis represents the number of pixels can be used. Specifically, the intensity distribution (shading) on the image may be binarized with a threshold value, or the intensity distribution may be linearized (contoured).

  According to the texture quantification apparatus of claim 1 and the texture quantification method of claim 4, the texture is identified based on the shape of the luminance histogram using the combined transformation correlation calculator, so that it is sufficiently accurate and high speed. Can be executed. In addition, analysis of surface roughness and gloss becomes easy.

  According to the texture quantification apparatus of the second aspect, since the joint transformation correlation calculation is performed using the same spatial light modulator and the same Fourier transformer, the optical system can be reduced in size.

  According to the texture quantification apparatus of the third aspect, since the joint conversion correlation calculation is performed using separate spatial light modulators and Fourier transform lenses, the correlation calculation processing time can be further increased.

1 is a block diagram schematically showing the configuration of Embodiment 1 of the present invention. (A) is a brightness | luminance histogram, (b) is explanatory drawing which shows the relationship between the distortion degree and glossiness. It is explanatory drawing which shows a distortion degree. (A) is a histogram when changing glossiness, (b) is a figure which shows the change of the degree of distortion when changing glossiness. (A) is a brightness histogram and glossy captured image when the gloss is changed, and (b) is a brightness histogram and surface roughness captured image when the roughness is changed. It is explanatory drawing which shows the technical principle of a joint conversion correlation calculator. It is explanatory drawing which shows a texture identification method. It is contrast explanatory drawing which shows experiment and simulation. It is a graph which shows an experimental result and a simulation result (gloss identification). 10 is a graph showing the fitting straight line (gloss identification) of FIG. 9. It is a graph which shows an experimental result and a simulation result (roughness identification). It is a graph which shows the fitting curve (roughness identification) of FIG. It is a graph which shows the comparison of the calculation time by a joint conversion correlation calculator and personal computer software. 6 is a block diagram schematically showing a configuration of a second embodiment. FIG.

  Hereinafter, preferred embodiments of a texture quantification apparatus and a texture quantification method according to the present invention will be described in detail with reference to the drawings.

  As shown in FIG. 1, the texture quantification apparatus 1 includes a laser diode 2 that outputs laser light (here, wavelength is 532 nm), an objective lens 3 with a magnification of 40 times, and a pinhole that uses laser light as a point light source. A search histogram, a collimating lens 5 (here, focal length f = 80 mm), a polarizer 6, a half mirror 7 and a texture quantification unit 13 described later, The input image in which the reference histogram is juxtaposed is input, spatial light modulation is performed based on the input image, and the spatial light modulator 8 which is a light image generating device for generating a light image, and the modulation from the spatial light modulator 8 A Fourier transform lens 10 (focal length f = 170 mm in this case) that receives light through a half mirror 7 and a polarizer 9 and optically Fourier transforms to form a Fourier transformed optical image; Includes a CMOS camera 11 for imaging (or a CCD camera), a light calculation correlation unit 12 configured as an optical computing correlator having a light image from the transform lens 10 as an output image. The texture quantification apparatus 1 is configured as a digital processing unit, and is an input image in which a search histogram that is set as a luminance histogram and is a search target and a reference histogram that is set as a luminance histogram and created in advance for collation are juxtaposed. And a texture quantification unit 13 that searches the search histogram by comparing it with a plurality of reference histograms. Here, the spatial light modulator 8 is a liquid crystal type spatial light modulator. For the texture quantification unit 13, the spatial light modulator 8 functions as an input screen, and the CMOS camera 11 functions as an output screen.

  Specifically, the texture quantification unit 13 specifically includes, for example, a CPU that executes various processes such as an image search process, a ROM that stores software programs necessary for processing operations, and an internal used for a reference histogram. And an input device, a display device, an external I / F, and the like. The input device is input means used for input necessary for processing operations such as image search processing in the texture quantification device 1. Examples of such an input device include a keyboard, a mouse, and various buttons and switches. The display device is configured by, for example, a liquid crystal display or the like, and is used for displaying data and texture necessary for the image search process and displaying the obtained image search result as necessary.

  As shown in FIG. 1, light having a wavelength of 532 nm is emitted from the laser diode 2, which is a coherent light source, and passes through the objective lens 3 and the pinhole unit 4 and collimated by the collimating lens 5. Is done. The collimated light passes through the polarizer 6 and the half mirror 7. The spatial light modulator 8 generates a reference histogram (referred to as a reference image b in FIG. 6) and a search histogram (referred to as a search image a in FIG. 6) at a distance ± d in the X direction from the center of the spatial light modulator 8. Display as input image. The light that has passed through the half mirror 7 is modulated according to the input image when reflected by the spatial modulator 8. Then, the modulated light modulated by the spatial light modulator 8 passes through the half mirror 7 and the polarizer 9, and reaches the Fourier transform lens 10, where it is Fourier transformed to generate a first light image, and the CMOS camera 11 The first converted image (interference fringe) is picked up and transmitted to the texture quantification unit 13.

  Next, as shown in FIGS. 1 and 6, the primary conversion image is input from the texture quantification unit 13 to the spatial light modulator 8 and displayed as an input image on the spatial light modulator 8. A second light image is generated by performing a Fourier transform for the second time, and is captured as an output image by the CMOS camera 11, and this output image is transmitted to the texture quantification unit 13. The correlation between the reference histogram and the search histogram is defined as a function of optical intensity at a point of distance ± 2d in the X direction from the center of the CMOS camera 9. In this way, the texture is recognized using the reference histogram and the search histogram set as the luminance histogram. Here, the optical calculation correlation unit 12 is a device that calculates correlation (similarity) between images by optical analog calculation, and requires downsizing of the optical system. VLC (matched filter) and JTC (joint conversion correlation) Can be considered. Although VLC can perform high-speed computation, it requires optical adjustment accuracy, requires two spatial light modulators, and increases the size of the optical system. JTC is easy to adjust optically and can be constructed with only one spatial light modulator, so that the optical system can be downsized. For this reason, in this embodiment, the JTC system is adopted. Since the optical computation correlation unit 12 includes one spatial light modulator 8, one Fourier transform lens 10, and one camera 11, the optical system can be reduced in size.

  This texture identification method is characterized in that the texture information is extracted as shape information. Gloss information is included in the shape of the luminance histogram. As shown in FIG. 2A, the luminance histogram is a two-dimensional image, the y-axis is the number of pixels, and the x-axis is the luminance (density). As shown in FIG. 2B, the degree of distortion of the mountains drawn on the black screen is related to the gloss. This degree of distortion is a measure showing the symmetry of the distribution of the number of pixels. When there is no gloss, the degree of distortion is 0.4, and when there is gloss, the degree of distortion is 1.4. Gloss tends to increase as the degree of distortion increases. As shown in FIG. 3 (a), the distribution having a long skirt on the left has a negative degree of distortion. As shown in FIG. 3 (b), the normal distribution (symmetrical) has a degree of distortion of 0, and FIG. As shown in (), a distribution with a long tail on the right has a positive degree of distortion. In FIG. 4A, the horizontal axis indicates gradation (luminance), the vertical axis indicates the number of pixels, and% indicates the degree of gloss (0% is not glossy and 100% is high gloss). However, as shown in FIG. 4B (showing the relationship between the gloss-containing area ratio (%) and the degree of distortion), the degree of distortion does not simply decrease as the gloss decreases. The degree of gloss cannot be determined. By supplementing other shape elements such as the shape of the upper protrusion of the luminance histogram, the degree of gloss can be determined.

  As shown in FIG. 5A, the shape of the luminance histogram changes according to the change in gloss. 0% is not glossy, 100% is high gloss, and the intermediate 25%, 50% and 75% indicate the degree of gloss. For identification of gloss, a surface image of a black paper with a rough surface coated with a gloss finish paint was used. On the other hand, as shown in FIG. 5B, the shape of the luminance histogram (such as the width of the skirt) also changes in accordance with the change in roughness (roughness). The surface images of sand cloths with different roughnesses were used to identify the roughness. Abrasive particles are present on the surface of the sand cloth.

In FIG. 5B, the number of abrasive particles per unit area (number of particles / cm ² ) is 40, 60, 80, 150, 240, and 400, respectively, and the larger the value, the smoother the surface state. The smaller the numerical value, the rougher the surface condition.

  Next, the cross-correlation calculation between the search histogram a and the reference histogram b performed by the light calculation correlation unit 12 will be described. As shown in FIG. 6, on the input screen of the spatial light modulator 8, the search histogram a (x, y) and the reference histogram b (x, y) are juxtaposed with an interval d, and with a focal length f. A Fourier transform lens 10 is disposed, and an output screen of the CMOS camera 11 is disposed at a focal distance f from the Fourier transform lens 10. The search image and the reference image on the input screen appear as interference fringes (primary transformed images) on the output screen. Next, when the interference fringes are input to the input screen and Fourier-transformed by the Fourier transform lens 10, an autocorrelation spot appears at the center of the output screen, and a cross-correlation spot appears at an interval 2d on the left and right. This is the output image.

  As shown in FIG. 7, the similarity of the shape of the luminance histogram depends on the glossy intensity of the cross-correlation spot. For this reason, the optical intensity at the cross-correlation spot is equivalent to the similarity of the texture. FIG. 8 shows the result of obtaining the interference fringes (primary transformed image) shown in FIG. 6 through experiments and simulations (FFT calculation).

  FIG. 9 shows the results when the gloss is identified by the experiment by the optical calculation correlation unit 12 and the simulation by the computer. Here, 100% (the most glossy) is used as a search histogram, and the search histogram is correlated with the reference histograms of 100%, 75%, 50%, 25%, and 0%. When the same histogram was used for the reference histogram and the search histogram, the cross correlation spot optical intensity was normalized to 100%. In both experiments and simulations, it was found that the gloss similarity decreased with decreasing gloss. The optical calculation correlation unit 12 was used for the experiment, and the simulation was performed using software, thereby calculating the optical intensity of the cross-correlation spot. As a result, in both experiments and simulations, gloss can be identified by the shape of the luminance histogram. As shown in FIG. 10, fitting (gloss identification) was performed using linear approximation, with the horizontal axis representing the percentage of area including gloss (%) and the vertical axis representing similarity. In the experiment, y = 0.92x + 1.49, and in the simulation, it becomes a linear function of y = 0.91x + 5.82, and the proportional constant ratio of the linear function between the experiment and the simulation is 0.91 / 0.92 × 100 = 99%. And the difference in proportionality constant is 1%.

FIG. 11 shows the results when the roughness is identified by the experiment by the optical calculation correlation unit 12 and the simulation by the computer. Here, 40 pieces / cm ² (the coarsest) was used as a search histogram, and a correlation calculation was performed between this search histogram and reference histograms of 40, 60, 80, 150, 240, and 400 pieces / cm ² . When the same histogram was used for the reference histogram and the search histogram, the cross correlation spot optical intensity was normalized to 100%. In both experiments and simulations, it was found that the roughness similarity decreases as the surface becomes smoother. The simulation was performed by software, and the optical intensity of the cross-correlation spot was calculated. As a result, in both experiments and simulations, it was possible to identify roughness based on the shape of the luminance histogram. As shown in FIG. 12, fitting (roughness identification) was performed using a power approximation with the horizontal axis representing the number of abrasive particles per unit area (number / cm ² ) and the vertical axis representing similarity. In the experiment, the exponential function was y = 3743X ^−0.98 , and in the simulation, y = 3115X ^−0.92 . The proportional constant ratio of the exponential function between the experiment and the simulation is 3115/3743 × 100 = 83%, the difference between the proportional constants is about 17%, and the exponent value ratio of the exponential function between the experiment and the simulation is 0.92 / 0.98 × 100 = 94%, and the index value difference is about 6%.

  As shown in FIG. 13, in the case of the experiment, since the optical calculation correlation unit 12 is used, the calculation time is constant regardless of the image size, but in the case of simulation, the calculation time is the image size because a calculator is used. Increases as becomes larger. Therefore, when the image size is large, the calculation time is shorter when the light calculation correlation unit 12 which is a combined conversion correlation calculator is used.

  By the way, regarding the luminance histogram described with reference to FIG. 2A, first, the surface layer of the search object is imaged by an imaging device (camera, scanner, etc.), and a luminance histogram is generated and recorded in the texture quantification unit 13. There are various methods for creating the luminance histogram. For example, in binarization, the lower side of the luminance histogram is set to white (“0”) and the upper side is set to black (“1”), or in the case of edging, a curve is drawn by line drawing ( “1”), and processing other than “0” is also possible. On the other hand, as the reference histogram, a plurality of brightness histograms created in advance are stored in a memory in accordance with the glossiness and surface roughness, stored in a database, and recorded in the texture quantification unit 13.

  Further, in the output image optically calculated by the JTC method, as shown in FIG. 7, the cross-correlation spot appears separately on both sides by the JTC method, and the auto-correlation spot appears in the middle. By detecting this pattern, the level of the correlation value is determined.

Here, a calculation formula equivalent to the optical correlation calculation using the JTC method for obtaining the autocorrelation spot and the cross-correlation spot will be briefly described. The reference histogram and the search histogram are a (x, y) and b (x, y), and these histograms that are separated from each other by + d and −d are juxtaposed as the input image f (x, y). x and y are coordinates. The input image f (x, y) is expressed by the following equation (1)
Formula 1: f (x, y) = a (x−d, y) + b (x + d, y)
Represented by If a function obtained by Fourier transform of the input image f (x, y) is F (Vx, Vy), this F (k) is expressed by the following equation (2).
Sought by.

For this function F (Vx, Vy), | F (Vx, Vy) | ² is obtained by using a complex conjugate function, and further converted into a power spectrum P in the real space. (3) is obtained.
Here, in the expression (3), the superscript ★ indicates a conjugate complex number.

In the right side of the equation (3), the first term and the second term correspond to autocorrelation spots in the correlation histogram. The third term and the fourth term respectively correspond to the cross-correlation spots that are separated by ± 2d from the autocorrelation spot.
Equation (4) shows the Fourier transform of the power spectrum P. In the formula, * indicates convolution and superscript ★ indicates a conjugate complex number. Therefore, the third term and the fourth term of the equation (4) are correlation data of a and b. Autocorrelation spots and cross-correlation spots are used as output images.

  What is characteristic in this embodiment is that the texture of an object is divided into roughness, gloss, and the like. Separation for obtaining correlation values can be performed by dividing into roughness, gloss, and the like. For the gloss, the degree of gloss can be found by identifying the protrusions at the top of the luminance histogram or various shape elements such as the degree of unbalance of the shape such as the degree of distortion, for example, the degree of extension to the brighter side. Gloss can be recognized by capturing the shape of the luminance histogram in the case of gloss. On the other hand, the surface roughness also varies depending on the shape of the luminance histogram (distribution width, etc.). For example, as the distribution width increases, the surface becomes rough. If such a shape pattern of the luminance histogram is registered in advance based on the shape of the luminance histogram, it can be matched at high speed according to the degree of correlation with it, and the gloss amount and the surface roughness can be quantified.

  The cross-correlation spot is calculated at high speed by the combined transformation correlation calculation, and this is digitized by the texture quantification unit 13. When digitized, the database can search for a brightness histogram that matches it.

  The similarity in texture is equivalent to the optical intensity of the cross-correlation spot. A large value indicates a high correlation. For example, various textures can be acquired using image search, such as arranging a plurality of reference histograms in descending order of similarity to the search histogram. A plurality of reference histogram patterns quantified in the database may be prepared, and the one with the highest optical intensity of the cross-correlation spot may be searched. The simulations shown in FIG. 9 to FIG. 12 are obtained by calculating the correlation value by performing the same operation with software on a computer. In a sense, it is close to true, with results close to experimental values.

  In this way, according to the configuration in which the original image is not used as it is, but the luminance histogram corresponding to the original image is referred to and the image search is performed by measuring the correlation degree of the texture using the optical correlation calculation by the JTC method. Image retrieval focusing on the texture can be executed at high speed with sufficient accuracy and by optical calculation.

  In addition, since JTC is used, the optical system can be downsized. For example, since the pixel pitch of the photodetector is fine (about 1.5 μm), the f value of the lens can be suppressed. However, it is necessary to increase the output of the laser diode 2 to some extent. In the matched filter, the f value of the lens is high and the size is increased, and accuracy is required for optical adjustment.

  Next, the texture quantification apparatus 101 according to the second embodiment will be described with reference to FIG. This is an embodiment that performs pipeline processing for performing joint transform correlation calculation at high speed. About the structure which is common in Embodiment 1, description is used as 100 series. The polarizer is not shown. The texture quantification apparatus 101 includes an optical calculation correlation unit 112 and a texture quantification unit 113. As shown in FIG. 14, light having a wavelength of 532 nm is emitted from a laser die auto 102 that is a coherent light source, passes through the objective lens 103 and the pinhole portion 104, and is collimated by the collimating lens 105. The collimated light passes through the half mirrors 107a and 107b, and irradiates a reference histogram and a search histogram displayed at a distance ± d in the X direction from the center of the spatial light modulator 108a. This display is performed based on the electrical signal from the texture quantification unit 113. The light reflected by the spatial light modulator 108a passes through the half mirror 107b, reaches the Fourier transform lens 110a, undergoes Fourier transform, and is imaged by the CMOS camera 111a. The optical signal is recorded as an input power spectrum by the CMOS camera 111a, and the recorded data is output to the spatial light modulator 108b at the next stage. On the other hand, the light reflected by the half mirror 107a is sent to the spatial light modulator 108b through the half mirrors 107c and 107d. Then, the light from the spatial light modulator 108b passes through the half mirror 107d, reaches the Fourier transform lens 110b, undergoes the second Fourier transform, and is imaged by the CMOS camera 111b. An image is captured as an input power spectrum by the CMOS camera 111 b, and the image is transmitted to the texture quantification unit 113. The correlation between the reference histogram and the search histogram is defined as a function of optical intensity at a point with a distance of ± 2d in the X direction from the center of the CMOS camera 111b. Although the apparatus is larger than that of the first embodiment by such pipeline processing, it is possible to perform the joint transform correlation calculation at a speed higher than that of the first embodiment.

  The present invention is not limited to the above-described embodiment, and various modifications and other embodiments can be adopted without departing from the technical idea of the present invention. For example, in the present embodiment, the gloss of an object due to light reflection (for example, diamond), the rough feeling due to light diffusion, the softness (file, cloth), and the like are recognized with high accuracy, but the present invention is not limited thereto. In addition, various kinds of texture such as transparency (for example, glass) of an object due to light transmission can also be targeted.

  Since the present invention quantifies surface roughness, gloss, etc. at high speed with high accuracy using joint transformation correlation calculation, it can be used for robots and the like.

DESCRIPTION OF SYMBOLS 1,101 ... Texture quantification apparatus 2,102 ... Laser diode 3,103 ... Objective lens 4,104 ... Pinhole part 5,105 ... Collimating lens 6 ... Polarizer 7, 107a, 107b, 107c, 107d ... Half mirror 8, 108a ... Spatial light modulator 9 ... Polarizers 10, 110a, 110b ... Fourier transform lenses 11, 111a, 111b ... CMOS cameras 12, 112 ... Optical calculation correlation units 13, 113 ... Texture quantification unit

Claims (4)

A spatial light modulator that modulates light from the light source, a Fourier transform lens that receives the modulated light from the spatial light modulator and performs Fourier transform, and a camera that captures an optical image from the Fourier transform lens, A light calculation correlation unit that performs a combined conversion correlation calculation of an input image input to the spatial light modulator and generates an output image;
The input image is set as a luminance histogram set as a search target to be searched and a reference histogram set as a luminance histogram and created in advance for collation, and the input image is set as the spatial light modulation. An output image including an autocorrelation spot and a cross-correlation spot obtained by the combined transformation correlation calculation is obtained from the camera, and the search histogram and the reference histogram are obtained based on the intensity of the cross-correlation spot. A texture quantification unit that collates and determines the similarity of textures;
A texture quantification device characterized by comprising:   A first modulated light is generated by the spatial light modulator to which the input image is input, a first optical image is generated by performing a Fourier transform of the first modulated light by the Fourier transform lens, and the first optical image is Imaging with the camera as a primary conversion image, outputting the primary conversion image to the spatial light modulator, generating second modulated light modulated by the primary conversion image with the spatial light modulator, and Fourier transform 2. The texture quantification according to claim 1, wherein a second light image is generated by performing a Fourier transform of the second modulated light with a lens, and the second light image is picked up by the camera and used as the output image. Device.   The optical correlation calculation unit includes a first spatial light modulator, a first Fourier transform lens that receives the first modulated light from the first spatial light modulator, performs a Fourier transform, and generates a first optical image; A first camera that captures a first light image from a 1 Fourier transform lens and generates a primary conversion image; a second spatial light modulator that inputs a primary conversion image from the first camera and generates a second modulated light; A second Fourier transform lens that receives the second modulated light from the second spatial light modulator and performs a Fourier transform to generate a second light image; and captures a second light image from the second Fourier transform lens, and The texture quantification apparatus according to claim 1, further comprising: a second camera that generates an output image. An input image in which a search histogram that is set as a luminance histogram and is a target for quantifying the texture and a reference histogram that is set as a luminance histogram and is created in advance is input to the spatial light modulator, and the spatial light modulator A first correlation calculation step of Fourier-transforming the light from the image by a Fourier transform lens to form a first light image, capturing the first light image and generating a primary transformed image;
The primary conversion image captured in the first correlation calculation step is input to a spatial light modulator, the light from the spatial light modulator is Fourier transformed by a Fourier transform lens to form a second light image, and the second light A second correlation calculation step of capturing an image to produce an output image including an autocorrelation spot and a cross-correlation spot;
A texture quantification step of collating the search histogram with the reference histogram based on the intensity of the cross-correlation spot of the output image to determine the similarity of the texture,
A texture quantification method characterized by comprising: