JP2010219865A - Image processing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique, by which preprocessing for each image recognizing section to perform image recognition processing can be omitted in an image processing system wherein an in-vehicle camera and a plurality of image recognizing sections are connected onto an in-vehicle network. <P>SOLUTION: In the image processing system, the in-vehicle camera on the in-vehicle network captures an image whose pixel value is represented by an RGB coloring system. A main component calculating section performs main component analysis for a captured frame image to determine a transform matrix. A pixel value transforming section acquires first to third main component scores from the transform matrix and the RGB pixel values of respective pixels, and distributes at least the first main component score to the in-vehicle network. An image recognizing section performs image recognition based on the first main component score with intensified contrast, and executes required subsequent processing. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image processing system in an in-vehicle network to which a plurality of image recognition units are connected.

  Currently, a camera is mounted on a vehicle to take a moving image or a still image. The captured image obtained by the in-vehicle camera has various uses, and is particularly useful for image recognition. For example, an accident can be prevented by detecting a vehicle, a pedestrian, a road surface condition, and the like from a photographed image and performing vehicle control or alerting the driver. In addition to image recognition, captured images are displayed on an image display unit in the vehicle and provided to the occupant, or stored in a storage device and recorded as driving records. When such an image recognition unit receives a captured image from an in-vehicle camera, image data is often distributed as a receiving node belonging to the in-vehicle network. That is, both the camera and the image recognition unit are connected to the in-vehicle network, and the captured image is distributed as digital data, and the image recognition unit receives and uses the data.

  Considering image data here, a moving image is composed of a plurality of frame images, and each frame image is composed of a predetermined number of pixels represented by RGB pixel values. However, if the pixel values of each frame are sent as they are for distribution, the amount of transmission data becomes too large, and the amount of network bandwidth consumed increases. In view of this, Japanese Patent Application Laid-Open No. 2004-104529 (Patent Document 1) introduces a method of compressing image data by the MPEG system or DV (digital video) system and distributing it to the in-vehicle network. Japanese Unexamined Patent Application Publication No. 2006-128766 (Patent Document 2) describes a technique for constructing an in-vehicle network at a low cost.

  The image data compression method described in Patent Document 1 uses a YUV color system as a color space. Here, it is known that there is a correlation between the parameters of the Y signal, the U signal, and the V signal. However, when the image recognition unit performs processing, the processing efficiency is better when the parameters do not have a correlation. Therefore, it is necessary to convert the image expressed in the YUV color system into an uncorrelated signal as preprocessing for image recognition. For example, principal component analysis has been used for this preprocessing, but processing delay and CPU consumption have increased due to an increase in calculation amount. In particular, when a plurality of image recognition units are connected to the in-vehicle network, it is necessary to perform preprocessing in each image recognition unit, and the processing is duplicated. Therefore, it is required to transmit uncorrelated data that does not require preprocessing in the image recognition unit.

  On the other hand, an image display unit such as a display is connected to the in-vehicle network to display a captured image. Also in this case, the network bandwidth can be saved by sending the RGB image after compression instead of sending it directly to the image display unit. However, if the format of the data sent to the image display unit and the data sent to the image recognition unit are different, the data must be created and transmitted separately at the transmission source, which increases the amount of network bandwidth consumed. . As a result, there is a possibility that convenience for the user may be reduced due to a delay in data transmission time or the like. In particular, when the transmission data is moving image data, since the amount of data is large, the above-described phenomenon is likely to occur. Therefore, when a captured image is distributed in an in-vehicle network in which the image recognition unit and the image display unit are connected together, it is necessary to make the format common to both.

JP 2004-104529 A JP 2006-128766 A

  The present invention has been made in view of the above circumstances, and an object of the present invention is to pre-process each image recognition unit in an image processing system in which an in-vehicle camera and a plurality of image recognition units are connected to an in-vehicle network. It is to provide technology that eliminates the need for A further object of the present invention is to provide a technique for suppressing consumption of a network band in an image processing system in which an image display unit is further connected to the in-vehicle network.

  In order to achieve the above object, the image processing system according to the present invention employs the following configuration. That is, an image processing system in which an imaging unit, a principal component analysis unit, a pixel value conversion unit, and a plurality of image recognition units are connected to an in-vehicle network to perform image recognition based on a captured image, and the imaging unit has a pixel value. Taking an image represented in the RGB color system, the principal component analysis unit obtains a conversion matrix from the photographed image, converts each pixel value of the photographed image into first to third principal component scores, The image processing system is characterized in that one principal component score is distributed to an in-vehicle network, and the image recognition unit receives the first principal component score and performs image recognition.

  In the conventional image processing system, pre-processing for acquiring data with strong contrast based on pixel values of the RGB color system is performed in each image recognition unit, but this is necessary by adopting the configuration of the present invention. Disappears. That is, since the image recognition unit uses the first principal component score distributed as the processing result of the principal component analysis unit, data suitable for image recognition can be obtained without performing preprocessing. Accordingly, since the plurality of image recognition units use the first principal component score in common, it is possible to suppress CPU consumption and improve processing speed. Further, since the performance of the CPU can be suppressed, the cost for constructing the image processing system can be reduced.

  In the image processing system according to the present invention, the principal component analysis unit also distributes the second and third principal component scores to the in-vehicle network, and the image processing system receives the first to third principal component scores and receives an image. The image reproduction unit for reproducing the image and the image display unit for displaying the reproduction image can be employed.

  By adopting such a configuration, the image recognition process and the process of restoring an image for use in image display can be performed based on the same conversion matrix. Therefore, there is no need to send RGB color system image data to the in-vehicle network for image display, and network bandwidth can be saved.

  In the image processing system according to the present invention, the image processing system further includes a feature amount calculation unit that calculates a feature amount representing screen brightness from the pixel value of the captured image, and the principal component analysis unit calculates a transformation matrix from the captured image. Each time it is stored, it is stored in the memory unit together with the feature amount of the photographed image from which the calculation is based. By using the conversion matrix that is used, each pixel value of the captured image can be converted into first to third principal component scores.

By adopting such a configuration, the principal component analysis is performed only when the brightness of the screen greatly changes, and if not, the previous conversion formula can be used. Therefore, the amount of calculation is reduced and the load on the CPU is reduced, so that the processing speed can be improved, parallel processing with other computations and replacement with an inexpensive CPU are possible. For example, an average value of pixel values of a captured image can be used as the feature amount.

  According to the present invention, in an image processing system in which an in-vehicle camera and a plurality of image recognition units are connected to an in-vehicle network, it is not necessary to perform preprocessing in each image recognition unit. Furthermore, in the image processing system in which the image display unit is connected to the in-vehicle network, it is possible to suppress the consumption of the network band.

1 is a block diagram illustrating a configuration of an in-vehicle network according to a first embodiment. 3 is a flowchart illustrating detailed processing according to the first exemplary embodiment. The block diagram which shows the structure of the vehicle-mounted network of Embodiment 2. FIG. 9 is a flowchart showing detailed processing of the second embodiment. FIG. 5 is a block diagram illustrating a configuration of an in-vehicle network according to a third embodiment. 10 is a flowchart showing detailed processing of the third embodiment.

  Exemplary embodiments of the present invention will be described in detail below with reference to the drawings.

<Embodiment 1>
In the present embodiment, a description will be given of a process in which each image recognition unit performs image recognition based on a captured image of an in-vehicle camera in an image processing system including two image recognition units.

(Characteristics of image recognition in vehicles)
First, since the present embodiment performs image recognition on a captured image of a vehicle-mounted camera, a brief description will be given. In addition, the characteristics of the image data to be recognized are also described. Image recognition is a type of image processing in which a program running on a CPU detects an object in an image on behalf of a person, makes a determination, and outputs the result. In particular, in a vehicle, image recognition having advantages such as certainty, endurance, and labor saving by performing a program is widely performed in order to increase safety. For example, there is a collision damage alleviating brake that applies a brake when a preceding vehicle is detected from an image taken in front of an in-vehicle camera, an inter-vehicle distance is calculated, and it is determined that the distance is too close. There is also a lane keep assist that detects a white line or a yellow line on a road from a photographed image, determines whether the vehicle is off the lane, or is likely to come off if it continues, and corrects the direction of the vehicle. In addition, processing that detects obstacles from the blind spot from the driver and notifies the driver with images and voices, detects text information such as road signs, and provides information to support driving Can be used.

  Image recognition proceeds in the following order. First, digitized image data is input to the image processing system, and feature quantities are extracted from the image data. Subsequently, the extracted feature quantity is compared with the standard pattern by pattern matching, and the object is recognized by the matching degree. The recognition result is output in a form usable by the user (or subsequent processing).

  Here, in the feature amount extraction of image data, in order to make it easy to recognize an object, processing such as detection of an outline or boundary line of an object in an image or extraction of an edge from a differential image is performed. Therefore, the feature quantity can be efficiently and accurately extracted as the change in color or density of the image to be processed becomes clear, that is, as the contrast of the image is stronger. As the accuracy of feature quantity extraction is higher, pattern matching becomes easier and the object becomes easier to recognize.

(Principal component analysis for RGB color system pixel values)
Next, since this embodiment performs principal component analysis on the RGB color system pixel values, a brief description will be given. Here, it is known that the RGB pixel values are redundant data having a relatively strong correlation with each other. Therefore, principal component analysis is performed on the pixel values of the RGB color system and converted to uncorrelated data. Principal component analysis is a technique for simplifying interpretation of data information by summarizing into a small number of uncorrelated variables when there is a correlation between multiple variables.

  The principal component analysis is performed, for example, by obtaining a variance-covariance matrix from a set of pixel values and obtaining its eigenvalue / eigenvector. Since the pixel value is three-dimensional (RGB) information, three sets of eigenvalues and eigenvectors are obtained. The eigenvector corresponding to the largest eigenvalue is the first principal component. Hereinafter, the second principal component and the third principal component are in order of the magnitude of the eigenvalue.

Based on the first to third principal components (eigenvectors), a conversion matrix or the like for converting the pixel values (r, g, b) into new pixel values (z ₁ , z ₂ , z ₃ ) can be obtained. it can. More specifically, first, a pixel set in which the average value of each color component is 0 by subtracting the average value of each color component obtained when obtaining the variance-covariance matrix from the pixel value of each pixel of the original image. Ask for. A new pixel value is obtained by creating a matrix from the eigenvector and multiplying the pixel value with an average value of 0. The new pixel values z _{1 to} z ₃ obtained by the conversion are called first to third principal component scores, respectively.

The first principal component score obtained by the above method has a large data distribution and a large amount of information. It can be said that this is a high contrast image suitable for image recognition. In addition, by performing an inverse transformation process of multiplying z _{1 to} z ₃ by the inverse matrix of the transformation matrix and adding the average value of each color component, the original pixel value can be obtained and the original image can be reproduced.

  In this way, by converting the RGB color system pixel values to the first to third principal component scores, uncorrelated data with no redundancy between the components can be obtained. Among the obtained data, the first principal component score has the largest amount of information. When viewed as an image, it can be said that the contrast is strong and the data can be recognized efficiently.

(Constitution)
The configuration of the image processing system used in this embodiment and the function of each block will be described with reference to the block diagram of FIG. Each block in the figure is mounted on the vehicle. The in-vehicle network 1 is a network installed in a vehicle, and each block is connected to enable data transmission / reception. The in-vehicle network is not limited to any constituent material or wired radio as long as data transmission is possible. For example, when the same data is distributed to a plurality of image recognition units as in this embodiment, multicast transmission can be made possible by using CAN, which is a standard.

  The photographing unit 2 is a digital video camera capable of photographing a moving image. In the present embodiment, the in-vehicle camera is installed on the dashboard facing the front of the vehicle, and images the traveling direction of the vehicle. An image is taken at a predetermined frame rate (for example, 30 frames per second), and each frame image has a predetermined number (for example, 640 × 400) of pixels having pixel values (for example, 0 to 255) represented in the RGB color system. It consists of The principal component analysis unit 3 includes a conversion matrix calculation unit 3a and a pixel value conversion unit 3b. The conversion matrix calculation unit obtains a conversion matrix for converting the coordinate axis based on the RGB pixel value of each pixel of the frame image. The pixel value conversion unit converts the RGB pixel values of each pixel of the frame image using a conversion matrix, and acquires first to third principal component scores. Then, the first principal component score used for image recognition is distributed to the in-vehicle network.

  The image recognition unit 4a receives the distributed first principal component score and performs image recognition. Here, the preceding vehicle is detected from the image, and the inter-vehicle distance is estimated by calculation. The brake control unit 5 receives the inter-vehicle distance, and performs brake control to avoid contact if the distance from the preceding vehicle is approaching. The image recognition unit 4b also acquires the first principal component score and performs image recognition. Here, an object indicating a road lane, such as a white line, a yellow line, or a guardrail, is detected from the image. The steering wheel control unit 6 receives the lane information, and corrects the course by performing steering wheel control if the vehicle is out of the lane or is likely to come off if the vehicle advances as it is.

(Processing flow)
Details of the processing will be described with reference to the flowchart shown in FIG. 2 as necessary. In step S101, the in-vehicle camera captures an image of 640 × 400 pixels at a rate of 30 frames per second as described above.

  In step S102, the pixel from which the transformation matrix is obtained is extracted from all the pixels of the frame image. This is because, depending on the performance of the CPU, if a conversion matrix is obtained based on all pixels, it may take too much processing time. Therefore, the number of pixels to be processed is reduced so that the image recognition process is not delayed or the CPU usage time is not increased and affects other processes. As a method for extracting pixels, in this embodiment, 1/2 downsampling is performed in the vertical and horizontal directions, and every other pixel is selected. As a result, the number of pixels to be processed can be reduced to ¼, and the processing can be speeded up. Whether or not to perform pixel extraction and the ratio of the number of pixels to be extracted to the total number of pixels can be determined according to the processing capability of the CPU and the accuracy required for image recognition. As a pixel extraction method, in addition to downsampling, a method of selecting pixels as many as the number of pixels necessary at random can be considered.

  In step S103, the conversion matrix calculation unit obtains a conversion matrix from the RGB pixel values of the sampled pixels. In step S104, the pixel value conversion unit converts the RGB pixel values of all the pixels of the frame image using the conversion matrix, and obtains first to third principal component scores. Since this pixel value conversion process is simpler than the principal component analysis, even if it is performed for all pixels, the amount of calculation does not increase. Subsequently, in step S105, the first principal component score is distributed on the in-vehicle network.

  In step S106, the image recognition unit 4a receives the first principal component score, performs image recognition, detects the preceding vehicle, and estimates the inter-vehicle distance. In step S107, the brake control unit determines whether the inter-vehicle distance is shorter than a predetermined value. If it is determined that the vehicle is too close to the preceding vehicle (S107 = Y), the process proceeds to step S108, where the brake control unit controls the brake and extends it to a safe interval. In addition, when the distance between vehicles is approaching but it is not enough to brake, the driver may be alerted by voice or the like. On the other hand, if S107 = N, the process proceeds to step S109, and the image recognition unit 5b detects road lane information. If the steering wheel control unit determines that the vehicle is out of or out of the lane (step S110 = Y), the process proceeds to step S111 to perform steering wheel control and correct the course. In addition, it is the same as the brake control that can be used as an alert to the driver instead of the steering wheel control.

(effect)
According to this embodiment, it is not necessary to separately perform preprocessing for obtaining an image suitable for image recognition by the two image recognition units. As a result, the CPU consumption is reduced, so that it is possible to construct an image processing system at low cost by suppressing the CPU performance of the image recognition unit. Further, the processing time in the image recognition unit is reduced, and subsequent processing can be performed with a small time lag. This makes it possible to improve the responsiveness particularly required for safety-related devices. In addition, since it is sufficient to distribute the first principal component score among the first to third principal component scores in order to perform image recognition, the data transmission amount is small compared to the case of distributing RGB color system pixel values, Bandwidth of the in-vehicle network can be saved. In addition, if the pixels from which the transformation matrix is obtained are extracted from all the pixels of the frame image by, for example, downsampling or the like, it is possible to obtain an advantage that the processing of the principal component analysis unit can be further speeded up.

<Embodiment 2>
In the present embodiment, each frame of a captured image is compared, and the principal component analysis process is performed only for a frame image whose image configuration has changed significantly (scene change has occurred). Explained.

  In each step of the process of the first embodiment, the principal component analysis is a process with a relatively large calculation amount and may become a bottleneck. Thus, in order to reduce the amount of calculation, it is conceivable to calculate the transformation matrix only for the selected frame instead of all the frames, and divert the previously calculated transformation matrix for the other frames. As a method for selecting a frame, for example, a simple selection at a predetermined interval such as every three frames may be considered. This selection method is effective in that the amount of calculation can be reduced efficiently.

  On the other hand, in order to reduce the amount of calculation while maintaining the accuracy of image recognition as much as possible, the presence or absence of a scene change is detected as in this embodiment, the transformation matrix is calculated from the frame where the scene change occurred, and the scene For frames that do not change, the method of diverting the previously calculated transformation matrix is effective. This is because there is a high possibility that the structure of the images are similar in adjacent frames in the moving image, so the RGB pixel values are also similar, and the conversion matrix obtained therefrom is considered to be similar. . In the present embodiment, it is determined that a scene change has occurred when the brightness of the frame image changes significantly.

(Constitution)
The system configuration of the present embodiment is shown in the block diagram of FIG. In addition to the first embodiment, the principal component analysis unit includes a feature amount calculation unit 7 that calculates an average value of the RGB pixel values of the frame image. In addition, a memory 8 exists for the principal component analysis unit to store the feature amount and the transformation matrix.

(Processing flow)
Details of the processing of this embodiment will be described with reference to the flowchart of FIG. In step S201, the in-vehicle camera captures an image. In step S202, the feature amount calculation unit calculates an average value of the RGB pixel values of the frame image. Since this calculation is simple addition and division, the amount of calculation does not increase even if all the frame images are processed. Hereinafter, the average value of the pixel values in the frame at the current time t is expressed as Pt.

  In step S203, the difference between the average value Pt of the RGB pixel values of the current frame and the value Pm stored in advance in the memory unit is taken. The value Pm is an average value of RGB pixel values in a frame image for which a conversion matrix has been obtained previously, and is stored at the time of principal component analysis.

  If the absolute value of the difference is equal to or smaller than the predetermined threshold value, the process proceeds to step S204 to instruct the pixel value conversion unit to use the conversion matrix stored in advance in the memory unit. This transformation matrix is obtained and stored when the principal component analysis was performed previously. On the other hand, if the absolute value of the difference is larger than the threshold value, the process proceeds to step S205 to obtain a new conversion matrix. Then, the newly obtained transformation matrix is stored in the memory unit. The threshold value can be set in consideration of the balance between the processing capability of the system and the accuracy of image recognition. Increasing the threshold value reduces the number of principal component analyzes and increases the processing speed, and reducing the threshold value enables principal component analysis. The number of times increases and the accuracy of image recognition increases.

  The subsequent processing is the same as in the first embodiment. In step S206, the pixel value conversion unit obtains first to third principal component scores from the conversion matrix and RGB pixel values determined in S204 or S205, and distributes the first principal component scores to the in-vehicle network. In step S207, the image recognition unit receives the first principal component score. In step S <b> 208, the image recognition unit performs image recognition processing such as estimation of an inter-vehicle distance by detecting a preceding vehicle and determination of lane deviation by detecting a white line or the like. In step S209, subsequent processing such as brake control and steering wheel control is performed using the image recognition result.

(effect)
According to the image processing system of this embodiment, it is possible to select whether a new transformation matrix is obtained or a transformation matrix stored in a memory is used after determining the presence or absence of a scene change. As a result, it is possible to reduce the number of calculations and suppress the CPU consumption while suppressing the influence on the accuracy of image recognition as much as possible. At this time, it is possible to select an appropriate threshold in consideration of the accuracy required for image recognition, the performance of the CPU, the processing speed, etc., and the degree of freedom in system design can be increased. If the number of principal component analyzes is reduced, the system can be configured with an inexpensive CPU to reduce costs. Alternatively, since the time required for the principal component analysis is reduced, it is possible to suppress the delay of processing and the influence on other processing.

  In addition to the absolute value of the difference, for example, a histogram of pixel values or a change rate can be used as the feature amount used in scene change detection. In addition, if information obtained from VICS (Vehicle Information and Communication System) is a trademark of the Road Traffic Information and Communication System Center, it is known that there is a scene change such as when exiting from inside a tunnel. A new transformation matrix may be obtained without obtaining a value.

<Embodiment 3>
In the present embodiment, not only image recognition but also a case where an original image is reproduced and used will be described. Hereinafter, the processing flow and the configuration of the image processing system will be described focusing on differences from the first embodiment.

(Constitution)
An image processing system according to the present embodiment is shown in FIG. 5, and components added to the first embodiment will be described. The image reproduction unit 9 acquires the first to third principal component scores sent to the in-vehicle network, performs inverse conversion processing, returns the pixel values of the RGB color system, and reproduces the original image. The image display unit 10 is a display device such as a liquid crystal display that displays a reproduced image. The image recording unit 11 is a storage device for storing reproduced images, and a hard disk or the like can be used.

(Processing flow)
The outline of the processing of this embodiment will be described with reference to the flowchart shown in FIG. In step S301, the in-vehicle camera takes an image. In step S302, the conversion matrix calculation unit obtains a conversion matrix from the captured frame image. Note that, in the process of obtaining the variance-covariance matrix at this time, the average value of each color component is obtained. In step S303, the pixel value conversion unit performs coordinate conversion to acquire first to third principal component scores. In step S304, the first to third principal component scores are distributed to the in-vehicle network. In the previous embodiment, it is sufficient to distribute the first principal component score used for image recognition. However, in this embodiment, since the image is also reproduced, the second and third principal component scores are also distributed. Furthermore, a conversion matrix and an average value of each color component necessary for reproducing an image are also distributed.

In step S305, the type of subsequent processing is determined and the processing is branched. When the subsequent process uses the image recognition result, the image recognition unit receives the first principal component score in step S306, and performs image recognition in step S307. In step S308, subsequent processes such as brake control and steering wheel control are performed.

  On the other hand, when the subsequent process uses the reproduced image, the image restoration unit receives the first to third principal component scores, the transformation matrix, and the average value of each color component in step S309. In step S310, the image reproduction unit performs a reverse process of the conversion performed by the pixel value conversion unit, and calculates RGB pixel values. In step S311, subsequent processing such as image display on the display and image storage in the storage device is performed.

(effect)
According to the image processing system of the present embodiment, image recognition and image reproduction can be performed using the first to third principal component scores sent to the in-vehicle network. In this way, by using the common data for the image recognition unit and the image reproduction unit, it is not necessary to send RGB color system image data to the in-vehicle network, and the network bandwidth can be saved. Further, since the image recognizing unit has a high contrast and can acquire the first principal component score suitable for image recognition, it is the same as in the first embodiment in that no pre-processing in the image recognizing unit is necessary. Therefore, especially in a system having a plurality of image recognition units, it is possible to eliminate duplication processing, and it is possible to reduce the cost of system construction.

  In the above embodiment, in order to simplify the process, the first to third principal component scores acquired by the principal component analysis unit are directly distributed to the in-vehicle network. However, the first to third principal component scores may be encoded in advance before distribution to provide encoded data with a compressed data size. In this case, the image recognition unit receives encoded data based on the first principal component score, performs decoding processing, and reproduces the first principal component score. The image reproducing unit receives all the encoded data and reproduces the first to third principal component scores. This reduces the bandwidth usage on the network and enables smooth data transmission. Alternatively, when data is stored in the image recording unit, the encoded data may be stored as it is without being subjected to a decoding process, and the image may be restored by decoding during reproduction. As a result, the storage area can be reduced and many operation records can be saved.

1 In-vehicle network 2 Imaging unit 3 Principal component analysis unit 4 Image recognition unit

Claims (4)

An image processing system in which an imaging unit, a principal component analysis unit, and a plurality of image recognition units are connected to an in-vehicle network and perform image recognition based on a captured image,
The photographing unit photographs an image in which pixel values are represented in an RGB color system,
The principal component analysis unit calculates a conversion matrix from the captured image, converts each pixel value of the captured image into first to third principal component scores, and distributes the first principal component score to the in-vehicle network;
The image processing system, wherein the image recognition unit receives the first principal component score and performs image recognition. The principal component analysis unit also distributes the second and third principal component scores to the in-vehicle network,
An image reproduction unit that receives the first to third principal component scores and reproduces an image;
The image processing system according to claim 1, further comprising an image display unit that displays the reproduced image. A feature amount calculation unit that calculates a feature amount representing the brightness of the screen from the pixel value of the captured image;
The principal component analysis unit further includes a conversion unit calculated from the captured image and a memory unit for storing the feature amount of the captured image that is the basis of the calculation,
If the difference between the feature amount of the photographed image and the feature amount stored in the memory unit is equal to or less than the threshold, the principal component analysis unit calculates each pixel value of the photographed image using a conversion matrix stored in the memory unit. The image processing system according to claim 1, wherein the image processing system converts the first to third principal component scores. The image processing system according to claim 3, wherein the feature amount is an average value of pixel values of a captured image.

Images