JP4280823B2 - Image numerical analysis method, system, and program - Google Patents

Description

  The present invention relates to an image numerical analysis method, system and program capable of expressing a color image with simple numerical values.

For the purpose of measuring and analyzing the amount of carbon dioxide absorbed by forests, it is often done to take pictures of changes in the landscape using a camera installed at a fixed point and record the plant season (phenology) as an image. . Usually, the detection of the event occurrence date and time using such image information is performed by comparing the individual images or making a moving image and visually observing the change. If this image information is represented by simple numerical values and a time series can be created, it is very convenient because it is easy to understand long-term fluctuations at a glance and to make yearly comparisons. In addition, various other numerical values measured at the same time as photographing, for example, combining image information as numerical values with meteorological elements, or using them for automatic processing of remote monitoring of ecosystems are also conceivable.
In image analysis using a conventional remote sensing method, the amount of radiation (brightness) of a specific wavelength measured in units of pixels using an expensive radiation meter, or the value obtained by converting the radiation into another index, and the pixel Create a correspondence table (teacher space) that correlates the object that contains the object, for example, whether it is a leaf of a tree or snow cover, theoretically from the radiation characteristics of the object, or the result of visual discrimination in the pattern of the image or in the field Then, all the pixels of the image are classified by this association. With this classification, it is possible to determine what a pixel whose subject is unknown is copied, or to obtain the area by counting the number of pixels for each classification over the entire image. However, this method involves complicated processing. In addition, the output of video equipment such as video cameras that are commercially available for general use is intended for viewing purposes only, and does not include information on the absolute value of the actual light energy, so precise analysis is required. Although it is not suitable for use, useful information may be extracted by simple analysis.

Since image information has a large amount of data, when used for remote monitoring, real-time monitoring of the image may be difficult due to restrictions on the amount of information transmitted and received depending on the communication means.
Therefore, an object of the present invention is to allow a low-speed communication means to collect a certain amount of useful information obtained from an image in real time by expressing a color image with simple numerical values.
Furthermore, it is intended to enable immediate application to numerical analysis by expressing complex image information with simple numerical indices.

  According to the image numerical analysis method and program of the present invention, a captured color image is stored for each pixel as digital image data obtained by separating the color into a plurality of colors for each pixel, and a single color for each separated color The average luminance value of a single color averaged over all pixels included in the image data and the average luminance value of all colors are obtained and normalized based on the average luminance value of all colors. The normalized average luminance value for each color is obtained. Create a time series of normalized average luminance values for each color, determine the magnitude of each value relative to 1.0 and the sign of the increase / decrease trend, and then determine the magnitude and increase / decrease trend of normalized average luminance value for each color relative to 1.0 The state of the corresponding color image is determined from the positive / negative combination pattern.

  The numerical analysis system for images according to the present invention comprises a field installation module and a data accumulation / analysis module. The on-site installation module captures a color image composed of a plurality of pixels and converts the color into digital image data obtained by separating the color into a plurality of colors for each pixel, a monochrome luminance value for each separated color, and all colors For the luminance value, the average luminance value of the single color averaged over all the pixels included in the image data and the average luminance value of all the colors are obtained, and the normalization for each color normalized based on the average value of the luminance values of all the colors A data collection and calculation unit for calculating the normalized average luminance value, and a communication unit for transmitting the calculated normalized average luminance value for each color. The data accumulation / analysis module creates a communication unit that receives the normalized average luminance value for each color transmitted from the field installation module, and creates a time series of the normalized average luminance value for each color. A data recording and calculation unit for determining the magnitude of the magnitude and the increase / decrease tendency with respect to 1.0, and determining the state of the corresponding color image from the pattern of a combination of the magnitude of the normalized average luminance value for each color with respect to 1.0 and the positive / negative tendency of the increase / decrease have.

According to the present invention, an image can be expressed by a simple numerical value, and thereby, even a low-speed communication means can collect some useful information obtained from the image in real time.
Furthermore, by expressing complex image information with simple numerical indices, it can be immediately applied to numerical analysis.
In recent years, digital video equipment such as digital still cameras has become available at low cost, and it has become possible to record image information simultaneously at multiple points, which increases the number of observation stations that record and store image information. is expected. Since it is not precise physical quantity measurement, quantitative analysis by remote sensing is difficult, but simple analysis as in the present invention is possible and useful information is expected to be obtained.

Hereinafter, the numerical analysis of the image of the present invention will be described by taking as an example the case of performing automatic plant season determination. In addition, the following terms used in this specification are used with the following meanings.
Plant season: Seasonal change in the state of vegetation (trees and herbs). For example, tree leaves open (extended leaves), autumn leaves, fallen leaves, etc.
Luminance value: Relative value recorded as an integer value on the recording medium after shooting by the video equipment, corresponding to “luminance” indicating the amount of light energy emitted from the subject.
Luminance value of a single color: Luminance value of a single color (red, green, and blue in RGB) for all colors Luminance value of all colors: Luminance value of a color (all colors, single color) Average luminance value: All colors Average value over all pixels in a single image of luminance value or single color luminance value Normalized average luminance value: A value obtained by normalizing the average luminance value of single color of each color based on the average luminance value of all colors

FIG. 1 is a diagram illustrating an algorithm for automatic plant season determination. In step ST1, images are taken every day at the same time (after noon), and are stored as digital image data for each pixel of a plurality of colors (hereinafter, RGB three colors will be described as an example) for separating the photographed color image. For each color, average luminance values r, g, b (monochromatic average luminance values for all colors) in all pixels included in the image data and average luminance values I of all colors in all pixels included in the entire image data are obtained. Obtained (ST2), values r _av , g _av , and b _av obtained by normalizing r, g, and b with I are obtained (ST3). Next, the magnitude for this r _av, g _av, to create a time series of b _av, a predetermined value of 1.0 for each value (average luminance value of a single color for each color is equal to the average luminance value of all colors), The sign of the increase / decrease tendency is obtained (ST4). Then, r _av, g _av, a pattern of positive and negative of the combination of large and small and increasing or decreasing trend for a given value of 1.0 b _av, determines the state of the corresponding forest (ST5). Hereinafter, each step of this algorithm will be further described.

(Step ST1: Shooting record)
First, shooting recording is performed. FIG. 2 is a diagram for explaining photographing by the camera. The example camera was installed slightly diagonally downward on a steel tower with a height of about 22 m. In order to capture a wider range, the lens is equipped with a wide-angle attachment. The area that can be included in the image is within a circle with a diameter of about 40 m on the ground surface. As shown in FIG. 3, an image captured by an image pickup device of a video camera is converted into an analog video signal (NTSC composite signal) by an encoder, and transmitted to a recording device installed under the tower through a signal line. The recording device comprises a personal computer and a video capture adapter (analog / digital converter for video signal), and converts the video signal received from the camera into digital image data and records it in a storage device (personal computer). . The resolution of digitized image data is 320 pixels wide and 240 pixels long. This landscape camera system places importance on reducing the storage capacity, and digitized images are recorded as data files compressed in JPEG format. For this reason, a certain amount of error (degradation of image quality) is caused by compression accompanied by image obscuration. In order to perform a precise analysis, it is desirable to record an image captured by the image sensor without being processed.

  In general, information obtained from digital video equipment is based on the three primary colors of light, R (red, red, generally 700 nm), G (green) , Green, generally 546.1nm), B (blue, blue, generally 435.8nm) monochromatic light, and the relative intensity of each monochromatic light (luminance value, most of which is 0 for each color) Is a three-dimensional array expressed as a set of integers from 255 to 255. The captured image used here is recorded with integer values of 256 gradations for each RGB.

(Step ST2: Obtain the average luminance value of each single color of R, G, and B and the average luminance value of all colors)
The simplest way to handle a distribution with a single number is to find an average value. Therefore, with respect to each day's image, the temporal change of the average value (monochromatic average luminance value) over the entire image of the luminance value for each of the R, G, and B single colors is obtained. R, G, B monochromatic average brightness value
Ask for. r _j , g _j , and b _j are the luminance values of single colors of R, G, and B in a certain pixel j, and n is the total number of pixels included in the entire image.

In addition, the amount of light incident on the camera lens is affected by the weather at the time of shooting, the solar altitude, and seasonal changes in the surface reflectance, etc. It has changed. A camera having an automatic exposure adjustment function controls the aperture and exposure time of a lens so that the brightness of a captured image is constant. The same adjustment is performed when the analog video signal is converted into digital data. The visual brightness (luminance value) i of pixel j is
It can be calculated with Cameras that adjust the exposure so that the visual brightness of the recorded image is kept constant change the coefficient of incident light intensity for each pixel depending on the position in the image (in many cameras, the coefficient increases as it approaches the center of the image). The weighted average value is adjusted and recorded as an index of the brightness of the entire image. For this reason, when a pixel with a large amount of incident light appears on a part of the image, the brightness of a relatively dark portion that covers the entire image may be recorded as an underestimated luminance value. Therefore, the simple average value of the luminance values of all colors over all pixels included in the entire captured image (average luminance value of all colors)
I asked again.

(Step ST3: Normalized by average luminance value of all colors)
In the photographed (Takayama site) image, the average luminance value of all colors was generally constant in the range of 153 to 204 in 256 gradations throughout the year, but there is some variation from day to day. To remove this variation in the average brightness value of all colors and to handle changes in the color of the forest ecosystem under constant light conditions,
Thus, values normalized by the average luminance values of all colors (normalized average luminance values) r _av , g _av , and b _av were used.

(Step ST4: Obtain an average color change with time)
A time-dependent change in the average color of the image is obtained from the time series of the values of r _av , g _av , and b _{av, and} the magnitude of each value, for example, 1.0 with respect to 1.0 and the sign of the increase / decrease tendency are obtained. In order to eliminate the influence of the shielding around the image, the range for obtaining the average value was calculated for 44000 pixels of 220 horizontal and 200 vertical in the center of the image as shown in FIG.

(Step ST5: Analyze changes over time and determine the state of the forest)
The analysis results are shown in FIG. The images collected and recorded every day at 14:00 were analyzed for two years from January 2001 to December 2002. Each year is shown in the upper and lower two figures. The broken line in each figure is a time series of r _av , g _av , and b _av . The color created by combining r _av , g _av , and b _av is equivalent to the average color of the entire image of each day, that is, the line in FIG. 5 expressed in color for each day (hereinafter referred to as the average color). To do. Roughly, in winter, r _av , g _av , and b _av are almost 1.0, the average color is gray, each color leaves the 1.0 line after April (around Julian day 90), and the average color is slightly Reddish. From the beginning of June (around Julius day 140) to October (around Julius day 270), g _av is the largest, and the average color is also green. After that, r _av was outstandingly large for about one month in October, and r _av , g _av , b _av converged again to the 1.0 line in November (Julius day, around 300 days) Both have the same seasonal changes. Intuitively, these changes correspond to winter snow cover white, ground and undergrowth brown or light brown after thaw, green leaves of summer trees, yellow and light brown leaves of fall and fall. it can. r _av is the tea color of soil, hay, autumn leaves and deciduous leaves, and g _av is easy to understand as the main color component of green color of undergrowth and leaves. B _av , that is, blue is difficult to understand as a color component of forests. However, considering the decrease in b _av as an increase in the yellow component that is its complementary color, it is easy to understand as the color contained in hay and tree leaves, autumn leaves and fallen leaves. Note that the b _av sometimes increases suddenly from 0.9 to 1.0 between spring and autumn, but this may be caused by fogging or condensation or frost formation on the camera lens cover. This is because. Although it is necessary to carry out detailed examination in the future, there is a possibility that this can be used to detect the occurrence of fog.

As described above, plant season events could be related to the values of r _av , g _av , b _av , and the characteristics of seasonal changes in color of the image average. These are summarized in Table 1. The relationship between the state of the ecosystem and the values of r _av , g _av , b _av appears as a level of value, especially a magnitude relative to 1.0, and the trend of its increase and decrease, and the ecological seasonal transition and its boundary are r _av , The increase / decrease tendency _{of the} values of g _av and b _av appears as a point where the change or the maximum and minimum values occur.

Table 2 shows the occurrence dates of plant seasonal events around the Takayama site from January 2001 to October 2003 estimated from the above analysis.

FIG. 6 is a diagram illustrating a plant season video telemetry system that executes the plant season automatic determination algorithm illustrated in FIG. 1. This system is composed of an on-site installation module and a data collection / analysis module. The on-site installation module includes an imaging unit, a data collection / calculation unit, and a communication unit. The imaging unit captures images every day at the same time (after noon) and saves them as digital (RGB) image data. The data collection and calculation unit calculates the average luminance values r, g, and b for the entire image data for R, G, and B, and the average luminance values for all colors for the entire image.
I is obtained, and r _av , g _av , and b _av obtained by normalizing r, g, and b with I are calculated. The communication unit transmits the calculated r _av , g _av , and b _av .

In the data accumulation / analysis module, the communication unit receives r _av , g _av , and b _av . Data acquisition, the arithmetic unit (1), the r _av, g _av, accumulates b _av, to create a time series. In this way, it is possible to visually determine the plant season of the forest, but the data recording and calculation unit (2) calculates the magnitude of each value with respect to 1.0 and the positive / negative of the increase / decrease trend, r _av , g _av , a pattern of positive and negative of the combination of large and small and increasing or decreasing trend with respect to 1.0 b _av, determines plant season corresponding forest. Hereinafter, this automatic pattern determination will be further described.

First, each day, 1 is subtracted from r _av , g _av , and b _av to obtain relative deviation values rr, rg, and rb from the average image brightness.
Calculate the time average of rr, rg, and rb for the previous 15 days including each day. Let this be rr, rg, rb. By treating rr, rg, and rb as moving average values, minute fluctuations (errors) caused by fog, snowfall, and weather at the time of shooting are removed, and seasonal trends are extracted.
For each of the moving average values rr, rg, rb, the difference from the moving average values rr, rg, rb 15 days ago is obtained. Let this be dr, dg, and db, and represent the increase and decrease trends of rr, rg, and rb 15 days ago.
A certain threshold value is set, and it is determined whether the values of rr, rg, rb, dr, dg, and db are positive, negative, or zero. In this example, the threshold values for rr, rg, and rb are ± 0.03, the threshold values for dr, dg, and db are ± 0.003. If the value is within this range, it is zero (0). Positive (+), and negative (-) if below this range.
The results of classifying the above six variables into three types (+,-, 0) are arranged in the order of rr, rg, rb, dr, dg, and db to form one pattern. For example, the pattern that appeared on the Julian day 90 in the figure of the analysis result is (+ 0- + 0-). Since rr, rg, and rb required on a certain day are values relating to 7 days before the current day, and dr, dg, and db are values relating to 15 days before the current day, it is necessary to create a pattern after combining the dates.

FIG. 7 shows the determination results for each day in 2002, and FIG. 8 shows 28 types of patterns that appear throughout the year. Each pattern is numbered from pattern 1 to pattern 28 in the order of the first appearance. As can be seen in the figure, the same pattern appeared repeatedly only during the midsummer, and the others could be classified as unique patterns by plant season (forest condition).
The correspondence between each pattern and the state of the forest determined by visually observing the image of the day is as shown in the figure. Patterns 12 to 19 in which the same pattern appears repeatedly are in a state where summer leaves are attached to the tree, and can be handled as a group.

  When the above method is used, there can be 3 6 = 729 patterns mathematically. Therefore, in the states other than the 28 types that appeared in 2002, the displayed patterns overlap with those of other seasons. If not, it can be added as a new classification after collating with visual discrimination of the image.

  FIG. 9 is a diagram for explaining the observation by the fixed point landscape camera system installed at the Takayama Observatory. The images used in this specification were taken at a carbon dioxide flux measurement station (Takayama site) set up by the National Institute of Advanced Industrial Science and Technology in a forest in Niugawa Village, Gifu Prefecture. The forest around the measuring station is a cold-temperate deciduous broad-leaved forest mainly composed of birch, quail, linden, birch, and maple wings. And there is a clear change, with branches, early summer leafs, dark green crowns in midsummer, autumn leaves and fallen leaves.

The seasonal change shown in FIG. 5 will be specifically described by taking 2002 as an example with few image defects. The fact that r _av , g _av , and b _av are all near 1.0 in winter means that most of the image is occupied by white to gray to black achromatic colors, which is due to snow on the ground surface. It is. Most of the trees in the analysis area are deciduous broad-leaved trees, so there are no leaves in the winter and all the ground surface under the tree branches is covered with snow.

On March 28, r _av , g _av and b _av started to change from around 1.0. This means that a saturated color has begun to appear in the image. As the snow thaws, the soil and undergrowth on the surface began to be exposed, and as the thaw progressed, r _av increased and b _av decreased. When r _av reaches the maximum, the snow on the surface is completely gone. During this time, the value of g _av is around 1.02 and has hardly changed since winter. In 2002, the thaw was completed on April 18, and r _av , g _av , and b _av were almost constant for about a month thereafter.

Around May 11th, tree leaves begin to spread, and on the image, the undergrowth of the ground is shielded. Along with this, g _av increases due to an increase in the area where the green color of the canopy appears, and r _{av of the} undergrowth relatively decreases. The increase in g _av continues until around June 1 when the leaves of all the trees in the image are complete. During this period, g _av increases and b _av decreases further compared to April, and takes a local minimum when leaf opening is complete. Since the color of the leaves of the tree immediately after spreading has a yellow component with a higher luminance than the color of the undergrowth, the decrease in b _av can be understood as an increase in the yellow component that is a complementary color as described above.

The entire image is covered with a canopy between July, August and September after the exhibition is completed, and g _av is outstanding. However, in detail, it can be seen that there is a gradual decrease with time. It can also be seen that r _av and b _av are increasing. This means that the average color is approaching an achromatic color. Since it is normalized by the average luminance value of all the colors of the entire image, it does not appear in the figure, but it is thought that this is because the leaves of the tree deteriorated and the color became dark as the season progressed.

At the end of September, the decrease in g _av accelerates, at the same time r _av increases and b _av begins to decrease again. This is because the color of the crown changed from green to yellow to red to light brown with autumn leaves. The average color also shows the reddish and yellowish color characteristic of autumn leaves. The start of the autumn leaves is difficult to identify the day because there is no clear bending in the line graphs of r _av , g _av , and b _av like other events, thaw and exhibition leaves. Melting snow and spreading leaves appear due to the increase or decrease in the area occupied by the white color of the snow or the green color of the leaves of the tree, whereas autumn leaves are due to the gradual change in the color of the subject itself. Conceivable.

The green leaves disappeared in the image around October 16, when g _av decreased to almost 1.0, r _av was exceptionally maximal, and b _av was minimal, and the autumn leaves reached their peak. Seem. At this point, littering begins, r _av decreases, and b _av increases. Looking closely at FIG. 4, it can be seen that the changes in r _av , g _av , and b _av associated with the peak season of autumn leaves do not occur all at once, but have a time difference. Since trees contained in the image have a approximately five, different colored leaves life and color by species, further also different for the height of the crown on species, of which r _av, g _av, b _av few days each change It seems that there is a difference. Observing the daily images of the autumnal leaves, the autumn leaves of linden trees began first, and the fallen leaves were completed on October 16. On the same day, yellow leaves of Dake birch and Mizunara, which occupy a large area in the image, reached the peak, and at the same time, defoliation began, and b _av reached a minimum. As replaced with this crown is colored leaves lower Kohauchiwakaede other trees, r _av is maximum reached after 3 days, it decreases with litter.

In 2002, there was snowfall on October 29, before the fallen leaves were completed, and the snowy season began on November 3, so the soil, undergrowth, and fallen leaves on the surface as seen in April. There was no exposed period, but until December 7th when the fallen leaves were completed on October 28 in 2001 and snow appeared on a part of the ground surface, the trees had no leaves and the entire image was soil on the ground surface. There was a period occupied by undergrowth, fallen leaves. During this period, as in April 2002, g _{av is} almost 1.0, r _av is slightly larger than 1.05, and slightly larger than the other two colors, and b _av is a constant value between 0.85, which is between the minimum and 1.0. I was taking it. After that, when the ground surface is covered with snow, r _av , g _av and b _av converge to around 1.0.

  A forest carbon dioxide balance telemetry system using a numerical image analysis method according to the present invention will be described with reference to FIG. 9 illustrating a fixed-point landscape camera system installed at a high mountain observatory. This telemetry system can always perform unattended monitoring of carbon dioxide absorption and release by forests. In connection with carbon dioxide emission reduction and emission trading, forests are thought to absorb and fix to reduce industrial carbon dioxide emissions. Under the Kyoto Protocol's operational rules, it was allowed to include absorption by forests, etc. as emission reduction targets. In order to calculate this amount of absorption, a scientifically supported measurement method is being studied. Also, at the research level for grasping the actual situation, it is necessary to monitor the amount of carbon dioxide fixed by forests at many points covering a wide area, and the development of a project to build a monitoring network is urgently required . Not only research-level projects that are currently underway, but also carbon dioxide emission trading including emission reductions due to forest conservation will be put into the implementation stage in the future. A means of monitoring at cost is required.

  The main monitoring items in this device are temperature, humidity, soil moisture content, light quantum amount, leaf area index, image information of phenology (landscapes such as tree leaves and fallen leaves, snow on the ground surface), and the like. Using these information, simple estimation of carbon dioxide absorption / release amount is performed. At observation points where precise measurement of carbon dioxide concentration (flux) is being conducted, the information collection and processing are also supported. Many of the places where the use of this apparatus is assumed are remote places where it is difficult to secure commercial power and arrange resident maintenance personnel. For this reason, electric power is generated by natural energy such as sunlight and wind power to operate the system. Similarly, automatic on-site processing of observation data, which can be easily handled even if the user is not proficient, and can reduce operating costs, and data is automatically collected at the data collection base using mobile phones and satellite communications Introduce telemetry technology such as The above system is packaged (product) in consideration of durability, price, and ease of operation. Eventually, data collected from monitoring package users is automatically converted into a database and networked for distribution over the Internet.

  The control and communication module controls all sensors attached to the observation site side package and acquires data, communicates with the data collection base, and further on-site processing of data (information from each sensor is logically integrated and It is a core device that plays a role in calculating carbon dioxide absorption and emission by the surrounding forest on the spot. These functions can also be realized using a normal personal computer. However, since it requires large electric power, it is difficult to drive with natural energy. In addition, the casing becomes large, and the degree of freedom in installation is lowered. Therefore, we applied the technology of “embedded computer” which has been attracting attention recently. In this apparatus, a sensor constructed on a single substrate is adopted in consideration of easy mounting of sensors, small power consumption and required volume, and easy housing miniaturization.

It is a figure which illustrates the algorithm of a plant season automatic determination. It is a figure explaining imaging | photography with a camera. It is a figure which illustrates the system which performs imaging | photography recording. It is a figure explaining the range which calculates | requires the time-dependent change of the average color of an image. It is a figure which shows the result of having analyzed the thing image | photographed and recorded at the fixed time every day for two years from January 2001 to December 2002. It is a figure which illustrates the plant season image | video telemetry system which performs the plant season automatic determination algorithm illustrated in FIG. It is a figure which shows the determination result about each day of 2002. It is a figure which shows 29 types of patterns which appeared throughout the year. It is a figure explaining the observation by the fixed point landscape camera system installed in the Takayama observatory.

Claims (3)

A captured color image consisting of a plurality of pixels is stored for each pixel as digital image data obtained by separating this color into a plurality of colors for each pixel,
With respect to the luminance value of a single color and the luminance value of all colors for each separated color, the average luminance value of the single color and the average luminance value of all colors averaged over all the pixels included in the image data are obtained, and the average of all the colors is obtained. Find the normalized average luminance value for each color normalized based on the luminance value,
Create a time series of normalized average luminance values for each color, determine the magnitude of each value relative to the predetermined value and the sign of the increase / decrease trend,
A numerical analysis method for an image, comprising: determining a state of a corresponding color image from a pattern of a combination of magnitude of the normalized average luminance value for each color with respect to the predetermined value and a positive / negative combination of increasing / decreasing tendency. In a numerical analysis system for images consisting of an on-site installation module and a data collection / analysis module,
The on-site installation module captures a color image composed of a plurality of pixels and converts the color into digital image data obtained by separating the color into a plurality of colors for each pixel; and
With respect to the luminance value of the single color and the luminance value of all colors for each separated color, the average luminance value of the single color averaged over all pixels included in the image data and the average luminance value of all the colors are obtained, and the luminance of all the colors A data collection and calculation unit for calculating a normalized average luminance value for each color normalized based on the average value,
A communication unit for transmitting the calculated normalized average luminance value for each color,
The data collection / analysis module includes a communication unit that receives a normalized average luminance value for each color transmitted from the field installation module;
Create a time series of normalized average luminance values for each color, and calculate the magnitude and increase / decrease of the normalized average luminance value for each color with respect to the predetermined value. A data recording and calculation unit for determining the state of the corresponding color image from a pattern of positive and negative combinations of trends,
A numerical analysis system for images. A captured color image consisting of a plurality of pixels is stored for each pixel as digital image data obtained by separating this color into a plurality of colors for each pixel,
The average luminance value of a single color averaged over all pixels included in each image data and the average luminance value of all colors are obtained, and the normalized average luminance value for each color normalized based on the average luminance value of all colors is obtained. Seeking
Create a time series of normalized average luminance values for each color, find the magnitude of each value relative to the predetermined value and the positive / negative of the increase / decrease trend,
An image numerical analysis program for executing each procedure for determining a state of a corresponding color image from a pattern of a combination of magnitude of the normalized average luminance value for each color with respect to the predetermined value and a positive / negative combination of increase / decrease tendency.