JP4781044B2 - Image analysis apparatus, image analysis method, and program for image analysis processing - Google Patents

Description

  The present invention relates to an image analysis apparatus and image analysis method for performing image analysis of a particle image obtained by photographing particles on a sheet, and a program for causing a computer to execute image analysis processing of a particle image obtained by photographing particles on a sheet. About.

  In recent years, particle analysis technology has been used for cells scattered in images acquired from a laser scanning microscope (total luminance value, peak value, area, major axis, minor axis, perimeter, etc.) There are optical cytometers that detect and present the statistics to the operator. The particle image generated by such an optical cytometer is likely to cause uneven brightness in the image due to unevenness in the amount of light due to the spot position of the irradiation laser, and this uneven brightness affects the particle extraction process and the feature amount of the particles. This could significantly reduce the statistical accuracy. Therefore, various countermeasures are taken in an apparatus that detects the characteristics of an object by image analysis such as the cytometer described above.

  A method for extracting low frequency components of an image with a filter and subtracting it from the original image as background information to solve inconveniences such as deterioration in feature detection accuracy due to luminance unevenness of the captured image data Alternatively, a quadratic surface equation is obtained by the least square method from the background position of the particle image visually confirmed by the operator and its luminance information, or from the automatically sampled background position and its luminance information, and is calculated from this quadratic curve. For example, a method of removing luminance unevenness by subtracting an image from the original image as background information has been considered.

  In addition, a method of averaging a large number of cross-sectional waveforms of luminance data in a certain direction of one original image and making the background uniform by the difference between the obtained waveform and each cross-sectional waveform of the original image, adding a large number of original images A method of creating an average image and making the background uniform by the difference between the addition average image and the original image is also common.

  Furthermore, for example, the coefficient of the background model function is numerically analyzed to obtain an optimum value that minimizes the square error of the background image created by the original image and the background model function, and the obtained coefficient is applied to the background model function. A method has been proposed in which a background image is created first, and an original image is calculated from the created background image to create a background removed image (see Patent Document 1).

JP 2002-92586 A

  However, the above-described method of creating a background image by averaging the cross-sectional waveforms and taking the difference from the original image has the advantage that a background image for removing the detected object from one piece of image data can be created. However, the background image cannot be generated by the averaging process especially when the ratio of the target object is large in the image data and the ratio of the background region is small. Another reason why the background image cannot be generated is that the particles that are the objects to be detected are periodically arranged.

  Further, in the example in which the background image is created by the averaging of a large number of images, the effect is large when the background unevenness is constant, but the effect is small when the background image is not constant and constantly changes. For example, it cannot be applied to feature detection for particles transported on a transport belt or the like. In addition, when the background image is non-uniform, there is a problem in that it cannot be binarized with the optimum gray scale uniformly over the entire screen.

  Therefore, the present invention has been proposed in view of such a conventional situation, and an image analysis apparatus that extracts an image of a measurement object over the entire screen by binarization from an image with uneven shading in the background. An object of the present invention is to provide an image analysis method and a program for image analysis processing.

In order to solve the above-described problem, the present invention eliminates a background image in an original image by performing a difference operation between the original image and a background image generated from the original image, thereby generating a particle-like shape from the original image with uneven density. an image analyzer for extracting an image of a detection object, an approximate image generating means for generating image data polynomial approximation in the horizontal direction of the original image waveform representing the brightness level for the pixels constituting the original image data, the approximate For each cycle in which the average level of the luminance level of the approximate waveform generated in the image generation means varies in the vertical direction of the original image, a pixel having the maximum or minimum luminance level of the approximate waveform in the vertical direction is selected. image processing said original image data by using a filter means for performing filtering, the image data that has undergone filtering by the filtering means Characterized in that it comprises an image processing means for extracting the image data of the detection object Te. Here, as the polynomial approximate image, it is particularly preferable to use an eighth order approximate image .

In addition, in order to solve the above-described problem, the present invention eliminates the background image in the original image by performing a difference operation between the original image and the background image generated from the original image, thereby removing particles from the original image with uneven density. In an image analysis method for extracting an image of a detection object in the form of an image, an approximate image generation step of generating image data obtained by approximating a waveform representing a luminance level with respect to pixels constituting the original image data in a polynomial approximation in the horizontal direction of the original image ; For each cycle in which the average level of the luminance level of the approximate waveform generated in the approximate image generation step varies in the vertical direction of the original image, the pixel having the maximum or minimum luminance level of the approximate waveform in the vertical direction is selected. a filter step of performing a filtering process of selecting, the detection object by image processing said original image data by using the image data that has been the filtered And having an image processing step of extracting the image data.

Further, in order to solve the above-described problem, the present invention is controlled by a computer and includes means for inputting original image data and means for performing predetermined image processing on the original image data, and is controlled by the computer. the information processing device the image analysis processing to extract the image of the particulate detection object from the original image is is executed, the program for executing a predetermined image analysis processing, for the pixels constituting the original image data an approximate image generation step of generating image data polynomial approximation in the horizontal direction of the waveform original image representing a brightness level, the brightness level average level of the approximate waveform generated in the approximate image generation process in the vertical direction of the original image every cycle of the periodic varying, filter the luminance level of the approximate waveform to select the maximum or minimum pixel in the vertical direction A filter step of performing management, and image processing the original image data using the image data that has been said filtering and image processing step of extracting the image data of the detection object, the original image and the original image The image processing apparatus is configured to execute an image analysis process for removing a background image in an original image by a difference operation with a background image generated from the image and extracting an image of a detection target object from an original image with uneven shading. To do.

  According to the present invention, when the ratio of the background region to the total area of the image is small, when the extraction objects are periodically arranged, and because the background image changes, the background image difference serving as a reference is calculated. Even if it cannot be determined, the background image can be efficiently removed and the image of the object can be extracted evenly.

  An image analysis apparatus shown as an embodiment of the present invention will be described in detail with reference to the drawings. The image analysis apparatus is an apparatus that performs image analysis for extracting feature quantities such as particulate materials and powder particles. Hereinafter, an image analysis device shown as an embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 shows a particle inspection apparatus to which an image analysis apparatus is applied. The particle inspection apparatus 1 mainly includes a granulation apparatus 2 that manufactures the particulate material 5, an observation system 3 that observes the manufactured particles 5, and an image analysis apparatus 4 that extracts feature quantities of the particles.

  The granulating apparatus 2 produces thermoplastic resin pellets or particles 5 such as powder particles and particulate materials, and discharges them onto a conveyor belt 31 extended to the observation system 3. Examples of the particles applicable to the present embodiment include methacrylic resins, methyl methacrylate-styrene copolymers (MS resins), polyolefins such as polyethylene (PE), polypropylene (PP), polycarbonate (PC), and polychlorinated. Particles of thermoplastic resins such as vinyl (PVC) and polystyrene (PS), particles of organic compounds such as fertilizer, and particles of inorganic compounds such as ammonium sulfate.

  The observation system 3 includes a conveyance belt 31 that conveys the particulate material or powder particles 5 produced by the granulation apparatus 2, and an observation window 32 for observing the particles 5 that are conveyed on the conveyance belt 31. Yes. As an example, the camera 41 is directed to the observation window 32. A CCD line sensor, a CCD area sensor, or the like can be used as the camera 41. Hereinafter, an example using a CCD area sensor will be described.

  The image analysis device 4 performs image analysis for extracting feature quantities of the particles 5 transported through the observation system 3. FIG. 2 shows the configuration of the image analysis device 4. The image analysis device 4 includes a CCD area sensor 41, an interface unit 42 for inputting image data acquired by the CCD area sensor 41, a memory driver 43 for controlling the image data to be stored in or read from the memory 44, An image processing unit 45 that performs image processing for calculating a background image from original image data acquired by the CCD area sensor 41 and stored in the memory 44 and extracting a detection target is provided. It is controlled and controlled by.

  The image analysis apparatus 4 further includes an operation panel 47 on which operation input is performed by the user, a display device driver 48 and a display device 49 for displaying messages and inspection results related to the inspection. The image analysis device 4 may be an information processing device such as a personal computer capable of executing image processing calculation.

  As an example, the CCD area sensor 41 is a sensor having about 512 × 480 pixels in the width direction, and light enters through a macro lens. Although not shown, the same sensors are arranged in parallel as appropriate to correspond to the width of the conveyor belt 31 (that is, the observation region). Further, it may be a laser generator for an optical cytometer, a scanning microscope, or the like.

  The interface unit 42 performs an A / D converter or the like on the image data sent from the CCD area sensor 41 to convert it into a data format handled in the subsequent image processing.

  The memory 44 temporarily stores the image data input from the interface unit 42. Or the image of the particle | grains as the detection target extracted by this Embodiment is stored. It may be used as a work area in a series of image processing.

  The image processing unit 45 generates a background image from the original image data and the original image data, and removes the background image in the original image by performing a difference operation between them to extract an image of the detection target object from the original image with shading. Image analysis is performed. FIG. 3 shows the configuration of the image processing unit 45. The image processing unit 45 is an input unit 441 that inputs original image data, and a polynomial approximation that generates image data obtained by polynomial approximation in the horizontal direction of a waveform that represents a luminance level for pixels constituting the original image data supplied from the input unit 441. Unit 442, a filter unit 443 that performs a filter process for selecting a pixel having the maximum or minimum luminance level of the approximate waveform, and image data of the detection target by performing image processing on the original image data using the filtered image data And a difference calculation unit 444 for extracting. The original images compared in the difference calculation unit 444 are supplied from the input unit 441 via the delay unit 445.

  In the original image, the background unevenness with respect to the particle 5 that is the detection object changes more slowly than the change period of the particle image that is carried by the conveyor belt 31. This indicates that the background curve can be expressed by a polynomial. However, the polynomial offset varies depending on the periodicity of the particle image. Therefore, a background image can be created if this variation can be removed after approximation by a polynomial. Therefore, in the present embodiment, the filter unit 443 performs a filter process for selecting a pixel having the maximum or minimum luminance level of the approximate waveform. In addition, although 8th-order expression was described as an example as a polynomial, the present invention is not limited to this. The image processing in the image processing unit 45 will be described in detail below with reference to the drawings.

  FIG. 4 shows an original image supplied to the input unit 441. The original image data of FIG. 4 is supplied to the polynomial approximation unit 442 and approximated by a polynomial. FIG. 5A shows a luminance level cross-sectional waveform and an eighth-order approximate deformation for a pixel on the line L1 passing through the center of the grain of the original image. FIG. 5B shows a cross-sectional waveform of the luminance level and an eighth-order approximate waveform for the pixels on the line L2 passing through the background of the original image. FIG. 6 shows an approximate image generated from a waveform obtained by approximating the cross-sectional waveform of the luminance level for the pixel of the original image data in FIG.

  The filter unit 443 performs a filter process for selecting a pixel having the maximum (or minimum) luminance level of the approximate image shown in FIG. 6 generated by the polynomial approximation unit 442. The result of the filter process is shown in FIG. The filter unit 443 supplies the filtered image data to the difference calculation unit 444. The difference calculation unit 444 calculates difference image data between the filtered image data shown in FIG. 7 and the original image data shown in FIG. Furthermore, the difference image data is binarized to extract image data of the detection target. The difference image is shown in FIG. 8, and the final binarized data of the particle 5 is shown in FIG.

  As described above, the image analysis apparatus 4 according to the embodiment of the present invention inputs the original image data captured by the CCD area sensor 41 and acquired through the interface unit 42, and the polynomial approximation unit 442 inputs the input unit. A filter that generates image data obtained by polynomial approximation of the waveform representing the luminance level for the pixels constituting the original image data supplied from 441 in the horizontal direction, and selects the pixel having the maximum or minimum luminance level of the approximate waveform in the filter unit 443 By performing processing and removing the background image in the original image by the difference calculation between the original image and the background image generated from the original image in the difference calculation unit 444 using the filtered image data, there is unevenness in light and shade A clear binary image of the detection object can be extracted from the original image.

  Note that the processing executed by each component in the image processing unit 45 is normally realized as a software module executed by the control unit 46. Each software module is executed by a program that executes a series of image analysis processes achieved by connecting each circuit configuration in FIG. 3 on a general multitasking OS having functions such as device control, a multitasking operation environment, and a timer. It can be realized by operating.

  In the following, the conventional (1) binarization method, (2) cross-sectional waveform addition average method, (3) addition average method of adding 20 original images, and (4) luminance with respect to pixels constituting the original image data Generate image data obtained by approximating a waveform representing a level in the horizontal direction by a polynomial (eighth-order equation), perform a filter process to select a pixel having the maximum or minimum luminance level of the generated approximate waveform, and obtain the filtered image data. An example in which image analysis is performed using each of the methods in the present embodiment in which original image data is processed and image data of a detection target is extracted will be compared.

  In contrast to the methods (1) to (4) above, (A) Extracted chunks extracted on one screen (substantially the same as the number of particles) (however, one block unit of 10 pixels or more connected areas on one screen) And (B) the average of the area occupied by the whole extracted mass, (B) the standard deviation, and (C) the maximum area occupied by the extracted mass. The results of the extracted particle images are shown in FIGS. 10 shows the original image data, and FIG. 15 shows the frequency distribution of the area occupied by the extracted block for each method.

  FIG. 11 shows the binarized data of the particle image calculated by the direct binarization method. At this time, there were 256 (A), (B) 96 pixels, (B) 5599, and (C) 89662. FIG. 12 shows binarized data of the particle image calculated by the cross-sectional waveform addition averaging method. At this time, there were 144 (A), (B) 1109 pixels, (B) 9490, and (C) 113200. FIG. 13 shows the binarized data of the particle image calculated by the addition averaging method in which 20 images are added. At this time, there were 256 (A), (B) 517 pixels, (B) 4454, and (C) 71335. FIG. 14 shows binarized data of particle images calculated by the method shown as the embodiment of the present invention. In the case of this embodiment, (A) 256, (B) 109 pixels, (B) 76, and (C) 501.

  Therefore, according to the image analysis apparatus 4 of the present embodiment, since a difference in luminance occurs between a cross section passing through the center of the particle and a cross section passing through a lot of backgrounds, a polynomial (here, an eighth order) approximate image is used. Therefore, it is possible to express only the background change while ignoring the luminance change due to the particles.

  In addition, since the average level of the approximate waveform varies greatly depending on whether the cross section passes through the particle or not, the image after the approximation becomes an image in which the luminance changes periodically. Therefore, this variation can be removed by performing a filtering process (minimum value filter, maximum value filter) that selects a pixel having the lowest (or higher) luminance within a range corresponding to one cycle.

  It was proved that background unevenness can be effectively removed by taking the difference between the background image generated in this way and the original image. Further, when looking at the binarized data of the particle image, the background removal effect is greatly improved as compared with the conventional binarized data of (1) to (3).

It is the schematic explaining the particle | grain inspection apparatus shown as embodiment of this invention. It is a block diagram explaining the image analysis apparatus in the said particle | grain inspection apparatus. It is a block diagram explaining the image processing part in the said image analysis apparatus. It is a figure which shows the original image acquired with the CCD area sensor. (A) is a waveform diagram showing a cross-sectional waveform of the luminance level for the pixels on the line L1 passing through the center of the grain of the original image and an eighth-order approximate deformation, and (b) is a line passing through many backgrounds of the original image. It is a wave form diagram which shows the cross-sectional waveform of the luminance level with respect to the pixel on L2, and an 8th-order type | formula approximate waveform. FIG. 5 is a diagram showing an approximate image generated from a waveform obtained by approximating the luminance level cross-sectional waveform with respect to the pixel of the original image data in FIG. It is a figure which shows the result of the filter process which selects the pixel whose luminance level of an approximate image is the maximum (or minimum). It is a figure which shows the difference image data of the image data shown in FIG. 7, and the original image data shown in FIG. It is a figure which shows the binarization data of the particle | grains finally extracted. It is a figure which shows original image data. It is a figure which shows the binarization data of the particle image calculated by the direct binarization method. It is a figure which shows the binarization data of the particle image computed by the addition average method of a cross-sectional waveform. It is a figure which shows the binarization data of the particle image calculated by the addition average method which added 20 images. It is a figure which shows the binarization data of the particle image calculated by the method shown as embodiment of this invention. It is the distribution map which showed the frequency distribution of the area occupied by the extraction lump for every method.

Explanation of symbols

  1 particle inspection device, 2 granulation device, 3 observation system, 4 image analysis device, 5 particles, 31 transport belt, 32 observation window, 41 camera, 42 interface unit, 43 memory driver, 44 memory, 45 image processing unit, 46 control unit, 47 operation panel, 48 display device driver, 49 display device

Claims (7)

In an image analysis apparatus that extracts an image of a particle-like detection target object from an original image with uneven shading by removing the background image in the original image by performing a difference operation between the original image and a background image generated from the original image.
Approximating image generating means for generating image data obtained by polynomial approximation of the waveform representing the luminance level for the pixels constituting the original image data in the horizontal direction of the original image ;
For each cycle in which the average level of the luminance level of the approximate waveform generated by the approximate image generating means varies in the vertical direction of the original image, the pixel having the maximum or minimum luminance level of the approximate waveform in the vertical direction is selected. Filter means for performing a filtering process to select;
An image analysis apparatus comprising: image processing means for performing image processing on the original image data using image data filtered by the filter means and extracting image data of the detection target.   The image analysis apparatus according to claim 1, wherein the polynomial approximate image is an eighth order approximate image.   The image processing means extracts the image data of the detection target by binarizing the difference image data between the image data filtered by the filter means and the original image data. The image analysis apparatus according to 1. In an image analysis method for extracting an image of a particulate detection target object from an original image with uneven density by removing the background image in the original image by performing a difference operation between the original image and a background image generated from the original image,
An approximate image generation step of generating image data obtained by polynomial approximation of the waveform representing the luminance level for the pixels constituting the original image data in the horizontal direction of the original image ;
For each cycle in which the average level of the luminance level of the approximate waveform generated in the approximate image generation step varies in the vertical direction of the original image, the pixel having the maximum or minimum luminance level of the approximate waveform in the vertical direction is selected. A filtering step for performing a filtering process to select;
An image processing method comprising: an image processing step of performing image processing on the original image data using the filtered image data to extract image data of the detection target. 5. The image analysis method according to claim 4 , wherein the polynomial approximate image is an eighth order approximate image. 5. The image processing step according to claim 4 , wherein the image data of the detection target is extracted by binarizing the difference image data between the filtered image data and the original image data. Image analysis method. A computer-controlled means for inputting original image data and a means for performing predetermined image processing on the original image data are provided, and an image of a particulate detection target object is extracted from the original image under the control of the computer In a program for executing predetermined image analysis processing on an information processing device that performs image analysis processing
An approximate image generation step of generating image data obtained by polynomial approximation of the waveform representing the luminance level for the pixels constituting the original image data in the horizontal direction of the original image ;
For each cycle in which the average level of the luminance level of the approximate waveform generated in the approximate image generation step varies in the vertical direction of the original image, the pixel having the maximum or minimum luminance level of the approximate waveform in the vertical direction is selected. A filtering step for performing a filtering process to select;
An image processing step of performing image processing on the original image data using the filtered image data and extracting image data of the detection target;
An image analysis process for removing the background image in the original image by the difference operation between the original image and the background image generated from the original image and extracting the image of the detection object from the original image with uneven shading is performed on the information processing device. A program characterized by being executed.

Images