Abstract: A leg (205) detection system comprising: a robotic arm (200) comprising a gripping portion (208) for holding a teat cup (203, 210) for attaching to a teat (1102, 1104, 1106, 1108, 203S, 203) of a dairy livestock (200, 202, 203); an imaging system coupled to the robotic arm (200) and configured to capture a first three-dimensional (3D) image (138, 2400, 2500) of a rearview of the dairy livestock (200, 202, 203) in a stall (402), the imaging system comprising a 3D camera (136, 138) or a laser (132), wherein each pixel of the first 3D image (138, 2400, 2500) is associated with a depth value; one or more memory (104) devices configured to store a reference (3D) 3D image (138, 2400, 2500) of the stall (402) without any dairy livestock (200, 202, 203); and a processor (102) communicatively coupled to the imaging system and the one or more memory (104) devices, the processor (102) configured to: access the first 3D image (138, 2400, 2500) and the reference (3D) 3D image (138, 2400, 2500); subtract the first 3D image

Abstract: The present disclosure discloses a multi-path image processing apparatus. An image merging circuit is configured to receive image frames that at least one of the image frames has a largest row number, generate redundant pixel row for each of the image frames that has a row number smaller than the largest row number such that the row number of each of the image frames equals to the largest row number, generate redundant pixel columns for each of the image frames having the number thereof determined by a size of a largest operation window, and merge each two of the image frames through the redundant columns thereof to generate a merged image frame. An image processing circuit performs image processing procedure on the merged image frame to generate a processed merged image frame, wherein at least a part of the image processing procedure is operated according to the largest operation window. An image segmentation circuit segments the processed merged image frame to generate processed image frames.

Abstract: Provided is a processing apparatus including a processing unit that is connected to a data bus, and performs control involved with an image which is output by each of a plurality of image sensors connected to the data bus, through the data bus.

Abstract: A method of performing an image retargeting quality assessment comprising comparing an original image and a retargeted image in a frequency domain, wherein the retargeted image is obtained by performing a retargeting algorithm on the original image. The disclosure also includes an apparatus comprising a processor configured to perform an image retargeting quality assessment, and compare an original image and a retargeted image in a spatial domain, wherein the retargeted image is obtained by performing a retargeting algorithm on the original image, and wherein comparing the original image and the retargeted image in the spatial domain comprises comparing the original image and the retargeted image to determine an amount of shape distortion between the images.

Abstract: Disclosed is an image creating device including a first obtaining unit which obtains an image including a face, a first extraction unit which extracts a face component image relating to main components of the face in the image and a direction of the face, a second obtaining unit which obtains a face contour image associated to the face in the image and a second extraction unit which extracts a direction of a face contour in the face contour image. The image creating device further includes a converting unit which converts at least one of the face component image and the face contour image based on the both directions of the face and the face contour and a creating unit which creates a portrait image by using at least one of the face component image and the face contour image being converted by the converting unit.

Abstract: A method and apparatus for tracking an object, and a method and apparatus for calculating object pose information are provided. The method of tracking the object obtains object feature point candidates by using a difference between pixel values of neighboring frames. A template matching process is performed in a predetermined region having the object feature point candidates as the center. Accordingly, it is possible to reduce a processing time needed for the template matching process. The method of tracking the object is robust in terms of sudden changes in lighting and partial occlusion. In addition, it is possible to track the object in real time. In addition, since the pose of the object, the pattern of the object, and the occlusion of the object are determined, detailed information on action patterns of the object can be obtained in real time.

Abstract: A method of embedding a binary watermark into an image and a computer applied such are disclosed. The binary watermark is transformed to a watermark sequence consisting of K binary numbers and the image is converted into a grayscale image, being segmented into several sections. K sections are selected randomly while each selected section corresponds to one of the binary numbers of the watermark sequence. In relation to each selected section, a sum of a first part of the pixels and a sum of a second part of the pixels is compared. Each of the selected sections having their result of comparison being made to correspond to corresponding binary number in accordance with a predefined correlation results in the binary watermark being embedded into the image.

Abstract: In various embodiments, methods, systems, and computer program products for processing digital images captured by a mobile device are disclosed. Myriad features enable and/or facilitate processing of such digital images using a mobile device that would otherwise be technically impossible or impractical, and furthermore address unique challenges presented by images captured using a camera rather than a traditional flat-bed scanner, paper-feed scanner or multifunction peripheral.

Abstract: Techniques are presented for analyzing audio-video segments, usually from multiple sources. A combined similarity measure is determined from text similarities and video similarities. The text and video similarities measure similarity between audio-video scenes for text and video, respectively. The combined similarity measure is then used to determine similar scenes in the audio-video segments. When the audio-video segments are from multiple audio-video sources, the similar scenes are common scenes in the audio-video segments. Similarities may be converted to or measured by distance. Distance matrices may be determined by using the similarity matrices. The text and video distance matrices are normalized before the combined similarity matrix is determined. Clustering is performed using distance values determined from the combined similarity matrix.

Abstract: An apparatus and method to find a specified number of corners and/or interest points in an image is presented. In some embodiments, a method to find a specified number of corners in a digital image comprises receiving a digital image containing a plurality of candidate corners; directly calculating a threshold score (S) for a center pixel, for each of the plurality of candidate corners, to form a plurality of scores; sorting, by the plurality of scores, the plurality of candidate corners to form a sorted list; and selecting corner locations sequentially from the sorted list based on the certain number of corners. The methods described may be implemented using any combination of hardware, software and firmware.

Abstract: A method for accurately measuring flicker, by first determining an original value based on a difference between pixel values in corresponding locations in a first picture and a second picture. Next, a reconstructed value is determined, based on a difference between pixel values in corresponding locations in a reconstruction from an encoding of the first picture and a reconstruction from an encoding of the second picture. Then, a determination is made to include a comparison value, which is determined on a difference between the reconstructed value and the original value, in a measurement of flicker between the first picture and the second picture.

Abstract: Methods and systems are provided for testing visual elements in a rendered web page. The method includes defining a gold image at a first point within a web application, taking a screen shot of an actual image at the first point during execution of the web application, and comparing the gold image to the actual image and generating a difference image based on the comparison. The difference image may include a first region highlighting a first difference between the gold image and the actual image within an area common to both images, and a second region highlighting a second difference between the gold image and the actual image which is not within an area common to both images.

Abstract: Restoration of edges in rasterized, continuous tone images includes selection of a pair of neighboring pixels that exchange a restoration value to make a steeper transition from a low value on one side of a boundary between the two pixels to a high value at the other side thereof. Although any pair of neighboring pixels is selected, this operation is only performed when four pixels in a direction perpendicular to the boundary, of which the selected pair form the two center pixels, have a monotonely increasing value. The restoration value is based on a difference value of nearest neighboring pixels on the same side of the boundary. This avoids overshooting around edges in an image. The method is advantageous when resealing an image to a higher resolution which occurs when a low resolution image is embedded in a page that is to be printed with a high addressability printer system.

Abstract: To compute a signature for an object comprising or represented by a set of vectors in a vector space of dimensionality D, statistics are computed that are indicative of distribution of the vectors of the set of vectors amongst a set of regions Ri, i=1, . . . , N of the vector space, at least some statistics associated with each region are binarized to generate sets of binary values ai, i=1, . . . , N indicative of statistics of the vectors of the set of vectors belonging to the respective regions Ri, i=1, . . . , N; and a vector set signature is defined for the set of vectors including the sets of binary values ai, i=1, . . . , N. The computing, binarizing, and defining operations may be repeated for two sets of vectors, and a quantitative comparison of the two sets of vectors determined based on the corresponding vector set signatures.

Abstract: When the number (M) of passes is smaller than a threshold value, a first processing mode is selected. In the first processing mode, multivalued image data is divided into pieces of multivalued data corresponding to passes and a common multivalued data for a plurality of passes, the pieces of multivalued data are individually binarized to generate pieces of binary data corresponding to the passes, and the common multivalued data is binarized to generate common binary data for these passes. On the other hand, when the number (M) of passes is equal to or larger than the threshold value, a second processing mode is selected. In the second processing mode, multivalued image data is binarized and the binary data is divided into pieces of binary data corresponding to passes with a mask.

Abstract: A motion compensated digitally subtracted Angiography (DSA) image processing system includes an interface for acquiring a sequence of images of patient vessels both prior to and following introduction of contrast agent into the vessels. An image data processor automatically, (a) determines a first shift vector for a first image of the sequence for compensating for shift between the first image and a first reference image of the sequence, (b) applies the determined first shift vector to the first image of the sequence to produce a shifted image, (c) subtracts the first reference image from the shifted image to produce a subtracted image enhancing vessel structure, (d) determines a second shift vector for compensating for shift between the subtracted image and a second reference image and (e) shifts content of the subtracted image relative to the second reference image in response to the second shift vector, to provide a shifted subtracted image enhancing and aligning vessel structure.

Abstract: A method and associated apparatus for topographically characterizing a workpiece. The workpiece is scanned with a scanning probe along a first directional grid, thereby scanning a reference surface and an area of interest subportion of the reference surface, at a variable pixel density including a first pixel density outside the area of interest and a second pixel density inside the area of interest to derive a first digital file characterizing topography of the workpiece. The workpiece is further scanned along the reference surface and the area of interest with the scanning probe along a second directional grid that is substantially orthogonal to the first directional grid and at a constant pixel density to derive a second digital file characterizing topography of the workpiece. A processor executes computer-readable instructions stored in memory that generate a topographical profile of the workpiece in relation to the first and second digital files.

Abstract: In accordance with one embodiment of the present disclosure, a difference is detected between a first image and a second image. The second image can include at least a portion of the first image reflected from a display panel and light from an object passing through the display panel.

Abstract: Techniques for capturing images of business cards, uploading the images to a designated computing device for processing and recognition are disclosed. A mechanism is provided to update extracted data from the images when there are any changes. Depending on implementation, there are a number of ways to capture images of business cards (e.g., via a phone camera, a PC camera, or a scanning device). A transmission means is provided to transport the images to the designated computing device for centralized management of integrated contact information for individual users. As a result, a user may access his/her updatable integrated contact information database anywhere anytime from a chosen device.

Abstract: A method for image-based weather condition detection, using image processing methods in moving vehicles, is described. The method includes detecting road weather conditions by identifying a group of bright pixels in video images produced by an image-based road characterization, and determining whether the bright pixels are caused by reflections due to one of rainfall, snowfall, a wet surface, or a snowy surface, where a driving assistant system includes a lane-keeping system.

Abstract: A system for image processing includes determining edge pixels of a model image using an edge based technique, and an angular orientation for each of the edge pixels of the model image. The system determines a lower spatial resolution model image based upon the model image and determining respective angular orientations for the lower spatial resolution model image, and determines edge pixels of an input image using an edge based technique, and an angular orientation for each of the edge pixels of the input image. The system determines a lower spatial resolution input image based upon the input image and determining respective angular orientations for the lower spatial resolution input image, and matches the lower spatial resolution model image with the lower spatial resolution input image to determine candidate locations of an object within the image and based upon the candidate locations matching the input image with the model image.

Abstract: An information retrieval apparatus includes an obtaining unit configured to obtain dynamic ranges or re-quantization codes of a target object to be retrieved and dynamic ranges or re-quantization codes of each of comparison objects to be compared with the target object, the dynamic ranges or re-quantization codes of the target object and the dynamic ranges or re-quantization codes of the comparison objects being obtained as a result of adaptive dynamic range coding; a distance computation unit configured to compute distances between the target object and the comparison objects using the obtained dynamic ranges or re-quantization codes of the target object and the obtained dynamic ranges or re-quantization codes of the comparison objects; and a comparison unit configured to compare the distances between the target object and the comparison objects to select one of the comparison objects having a minimum distance.

Abstract: A computer implemented method of determining a view impact of a target structure, the method constituted of: loading a 3D scene; determining a view point and direction of interest of the loaded 3D scene; rendering, without the target structure, a plane of the loaded 3D scene image perpendicular to the determined view point and direction of interest at a distance from the view point distal of the target structure; counting the number of pixels for the rendered plane without the target structure; rendering, with the target structure, a plane of the loaded 3D scene image perpendicular to the determined view point and direction of interest at the distance from the view point distal of the target structure; counting the number of pixels for the rendered plane with the target structure; and calculating the difference between the counted pixels with the target structure and the counted pixels without the target structure.

Abstract: A method which uses digital image comparison for imaging software development is described. The software under development is used to generate a set of test digital images from print data. The images are stored. The software of a reference system is used to generate a set of reference digital images based on the same print data. The test and reference images are tiled, and image comparison is carried out on a tile-by-tile basis. A difference tile is generated for each test image tile and corresponding reference image tile, and the tiles are stored together in an image file to be displayed to the human user. The test images and reference images are compared using an image comparison program. The result of the comparison is presented to the human software developer for further comparison and evaluation.

Abstract: Computer-implemented image processing methods and apparatuses are presented for automatically selecting regions of interest within an image represented by pixel intensity values. A first pixel box is employed in progressively scanning and evaluating the image. If pixels within the first pixel box have pixel-intensity-related characteristics exceeding respective defined thresholds, then those pixels are identified as an area of interest and a second pixel box is employed in progressively scanning and evaluating the selected area of interest to identify regions of interest. Each area of interest is larger than a region of interest, and the second pixel box is smaller than the first. Regions of interest within the image are identified if one or more pixel-intensity-related characteristics of pixels within the second pixel box exceeds a second defined threshold, wherein the second threshold is greater than the first. Once selected, identifying information for the regions of interest is stored or output.

Abstract: A method, system and devices for optimizing and/or maximizing the perceived autostereoscopic effect for any chosen combination of photographic or electronic recording materials, recording methods and parameters, and imaging methods and parameters. The method first determines the parameters characterizing the optimum autostereoscopic effect of the combination materials etc. (the front pixel spread and the back pixel spread), then specifies how to apply these parameters to the arrangement of image acquisition devices, or the production of images by computational means, to produce a final image with optimal autostereoscopic properties. Features of the present invention provide for image acquisition either occurring in the real world (via digital or analog means) or within a computer graphic environment (or a blend of both), by way of specific methods necessary for successfully traveling through a variety of different mathematical spaces to arrive at a common space.

Abstract: Software failures are processed by categorizing images associated with the failures into groups based on similarities in the images. The categorizing of the images may be performed by a software failure analysis tool. The images may include full or partial screenshots of a computer at the time of the software failure. An image may be part of the output of a test, or may be part of a report of a software failure from an end user of the software. The software failure analysis tool may be part of a larger software test management system.

Abstract: A similar image having a high correlation is selected through autocorrelation performs a template original image selected from an image photographed for a template, and a difference image between the similar image and template original image is formed. An image extracting a real difference is formed by removing noises and edges in unstable areas from the difference image. This image is added to the template original image to form a modified template. Template matching is performed by using the modified template as a template. The image extracting the real reference and added to the modified template functions to add an evaluation penalty to the similar image during matching evaluation to lower an evaluation value of the similar image so that a probability of erroneously recognizing the similar image as the image to be detected.

Abstract: The present disclosure includes systems and methods relating to digital image modification using pyramid vignettes. In general, one aspect of the subject matter described in this specification can be embodied in a method that includes receiving a request associated with rendering a photorealistic variation of a master digital image at an original resolution, selecting a digital image template from a number of digital image templates comprising image data associated with the master digital image at various resolutions, the selected image data at a different resolution from the original resolution, and rendering the photorealistic variation of the master digital image using the selected template.

Abstract: A system, method, and computer-readable media are presented for presenting similar images of a selected image in a user interface. One embodiment of the method can include identifying a selected first image and identifying a selected direction from the first image. Additionally, the method can include calculating a best neighbor metric for each of one or more other images to determine a best neighbor image in the selected direction. Furthermore, the method can include presenting the best neighbor image to a user in a user interface.

Abstract: A method for contour reduction in a digital picture is disclosed. The method generally includes the steps of (A) buffering a plurality of luma samples in a current line of the digital picture, each of the luma samples having a respective input value, (B) calculating a plurality of horizontal sum-of-signs along the current line, wherein each of the horizontal sum-of-signs comprises a sum of a plurality of amplitude differences between pairs the luma samples from the current line and (C) generating a plurality of output value based on the horizontal sum-of-signs, one of the output values for each one of the luma samples.

Abstract: This invention consists in an image processing method in which the similarity between a first pixel and a second pixel constituting an image is numerically given by statistical testing, and the values of the first and second pixels are averaged when the similarity numerically given is high, whereas the first and second pixel values are not averaged when the determined similarity is low, whereby an image with noises suppressed is obtained without spoiling a spatial resolution as well as a temporal resolution.

Abstract: A method for accurately measuring flicker, by first determining an original value based on a difference between pixel values in corresponding locations in a first picture and a second picture. Next, a reconstructed value is determined, based on a difference between pixel values in corresponding locations in a reconstruction from an encoding of the first picture and a reconstruction from an encoding of the second picture. Then, a determination is made to include a comparison value, which is determined on a difference between the reconstructed value and the original value, in a measurement of flicker between the first picture and the second picture.

Abstract: A system and method for detecting differences in images and identifying groups of images with similar differences detected is provided. To this end, an image comparator may be operably coupled to a pixel comparator for comparing pixels of the baseline image with corresponding pixels of one or more sample images to detect differences between the pixels compared. A derivative difference processor may be also provided which may be operably coupled to the pixel comparator for determining a difference region of pixels detected as different by the pixel comparator in the sample images. An image sorter may also be provided which may be operably coupled to the difference derivative processor for grouping together sample images that have a similar difference region determined by the derivative difference processor.

Abstract: Output from a video camera is monitored. Input data is analyzed to produce a model of a background image and incoming image data is compared with said background model to identify images having a high level of activity created by the introduction of a foreground object. A period of activity, composed of a plurality of consecutive images having a high level of activity, is identified, and an exemplar image is generated.

Abstract: A method is provided that increases throughput and decreases the memory requirements for matching multiple templates in image. The method includes determining a set of inter-template early elimination values that characterize the degree of matching between various templates and the image, at various locations in the image. A later-analyzed template may be rejected as a potential match at a location in the image based on comparing a value characterizing its degree of match at that location to an inter-template early elimination value corresponding to the degree of match of an earlier-analyzed template at that location. The compared values may be determined by different sets of operations, and may be normalized such that they are properly comparable. The inter-template early elimination conditions may be stored in a shared correlation map. The shared correlation map may be analyzed to determine the matching locations for multiple templates in the image.

Abstract: An exemplary method for estimating pixel values of displayed images captured by an image capture device comprises determining a status of a pixel of a displayed image based on a confidence value for the pixel and a comparison of an estimated value and a direct sampling value of the pixel, determining whether the pixel is obstructed based at least in part on the status and the confidence value, determining whether to update a pixel value of the pixel based at least in part on the status and whether the pixel is obstructed, repeating the determining steps for other pixels of the displayed image.

Abstract: Methods, apparatuses, and systems are presented for measuring difference between graphics images relating to performing an arithmetic operation involving a first graphics image comprising a plurality of first pixels and a second graphics image comprising a plurality of second pixels to produce a difference image comprising a plurality of difference pixels, generating measures of proximity from a plurality of ranges of observation within the difference image, wherein the measures of proximity represent spatial proximity of difference pixels to other difference pixels within the difference image, and applying non-uniform weighting to the difference pixels to produce a weighted difference image, wherein the non-uniform weighting depends on the measures of proximity generated from the plurality of ranges of observation.

Abstract: A vehicle restraint system has a vision-based occupant classification system for control of airbag deployment during a crash scenario. The classification system utilizes two imaging sensors which together create a stream of paired images received and stored by an occupant classification controller. A computer program product of the controller utilizes the paired images to extract disparity/range features and stereo-vision differential edge density features. Moreover, the controller extracts wavelet features from one of the two paired images. All three features or maps are classified amongst preferably seven classifications by algorithms of the computer program product producing class confidence data fed to a sensor fusion engine of the controller for processing and output of an airbag control signal input into a restraint controller of the vehicle restraint system.

Abstract: A system and method for improved image comparison is provided. An adjustable tolerance level may be set for comparing the images, and filtering may be applied to the images, for example, to suppress high frequencies in the images or to enhance low frequencies in the images. Furthermore, a geometrical transform, such as a 2D affine transform, may be applied to one of the images. The pixels of one of the images may be compared with corresponding pixels of another image to detect differences between the pixels compared. Then the outcome of comparing the pixels of the images may be determined by comparing the differences detected with the tolerance level set. The tolerance may be set by selecting one of a set of predefined tolerance levels, each of which may represent a different degree of tolerance to be used in comparison of the images.

Abstract: Methods and apparatus, including computer program products, for classifying elements of a digital image. An element of the digital image and a prototype representing a class of elements are received and a difference image is generated. The difference image represents differences between the received element and the prototype and includes ON pixels that represent local differences between the received element and the prototype. The difference image includes one or more clusters of neighboring ON pixels. Relative arrangements of ON pixels are determined in the difference image. One or more scores are assigned to the difference image using the relative arrangements of ON pixels in the difference image. Based on the assigned scores, it is determined whether the received element belongs to the class represented by the prototype.

Abstract: An exemplary method includes identifying defects in a first image among a series of images, by identifying a first portion of the first image at a first location in the first image, wherein a brightness of the first portion is different from a second portion at the first location in a previous image and a third portion at the first location in a subsequent image, by an amount exceeding a first threshold value, determining whether a first region of the first image that includes the first portion, matches a second region of the previous image and a third region of the subsequent image, wherein the second and third regions are at second and third locations displaced from the first location, and characterizing the first portion as a defect when the first region does not match the second and third regions.

Abstract: A system and method for estimating noise using measurement based parametric fitting non-uniformity correction is disclosed. Fixed pattern noise (“FPN”) is estimating from an overall noise component within a detection system to enhance candidate target detection and tracking. A sensor in the detection system receives energy, such as radiant flux, that is converted to a digital image. A non-uniformity correction device generates an estimated FPN according to an applicable temperate range and integration time. A memory storing an array of coefficients is accessed to determine the estimated FPN. The valves within the array of coefficients are based on actual FPN measurements that are parametrically fitted.

Abstract: A target recognizing device that creates a template concerning a target by itself is provided. The device comprises a robot camera for creating continuous images showing a target and an arithmetic operation unit for controlling the robot camera. A robot camera acquires continuous image information by means of a camera rotatable by a motor. The arithmetic operation unit comprises a template management section for allowing an update part to update a template stored in a template storage part and a target recognition section having an image storage part where the acquired image information is stored, a recognizing part for recognizing a target by using the template, and a motion command part for rotating a camera according to the result of recognition.

Abstract: Diminished intensity defects occur in electrostatic printing between image regions having grey levels, i.e., different electrostatic potential and toner densities. Such defects occur when higher density regions “steal” toner from lower density regions. The system and methods according to this invention compensate for these defects by modifying the input image data. The input image data in lighter regions that precede or occur near a light-to-dark transition to a dark region are raised above the input image values. Thus, when printed, the printed image intensity values in such regions are higher than the corresponding image intensity values. As a result, when the higher density regions steal the extra toner provided due to the raised values, the gray level of the printed image corresponds to the desired gray level. The magnitude of these defects is periodically measured with a calibration pattern to generate and/or update compensation factors used in the compensation process.

Abstract: A method for extracting a watermark signal contained in a watermarked digital image sequence having two or more frames, including the steps of estimating correspondences between one or more pairs of frames in the watermarked digital image sequence; computing a displaced frame difference for one or more frames in the watermarked digital image sequence using the correspondences; and extracting the watermark signal from one or more displaced frame differences.

Abstract: Methods and systems provide automated testing of computer-generated displays. The proper functionality of a memory storage device on a computer video card and the proper functionality of software for generating computer-generated displays may be tested by storing a display image to a first memory device context while displaying the same image on a computer screen viewable by a user. The image displayed to the computer screen is captured into a second memory device context. The image in the first memory device context and the memory in the second device context are compared on a pixel-by-pixel basis to determine whether the two stored images match. If the second stored image does not match the first stored image, an indication is presented that the video memory of the computer memory card does not operate properly or that software responsible for displaying the image to the computer display screen is not operating properly.

Abstract: A method for matching digital images, including regularization of image features of a first digital image, composed of pixels, defining a finite set of candidate values, wherein a candidate value represents a candidate for a possible match between image features of the first image and a second image, establishing a matching penalty function for evaluation of the candidate values, evaluating the matching penalty function for every candidate value, selection of a candidate value based on the result of the evaluation of the matching penalty function, regularization of the first image by segmentation of the first image, comprising assigning at least part of the pixels of the first image to respective segments, determining a certainty parameter for at least part of the pixels of a segment, and establishing the matching penalty function to be at least partially based on the certainty parameter.

Abstract: A method and apparatus for measuring a color-texture distance and a method and apparatus for sectioning an image into a plurality of regions using the measured color-texture distance are provided.

Abstract: A marginal distribution feature extraction unit 20 sequentially focuses on each pixel in a reference image and the corresponding pixel in an inspection image, and computes a marginal distribution feature value indicative of the spatial variation in pixel values in the focus pixel neighborhood in both images. Based on this marginal distribution feature value, a tolerance image generation selection unit 26 sets a tolerance range for the reference image or inspection image focus pixel with the less spatial variation. A target image selection unit 30 selects the image comprising pixels from both images for which a tolerance range is not set as a target image. Referencing the set tolerance ranges, a comparison operator 34 compares each pixel in the target image and tolerance image, and outputs a difference Sub representing how far the pixel values of the target image are from the respective tolerance range.