Abstract: A method and system for recognizing an object in an image. A model of the sought object is transformed according to a set of preselected transformations to produce a set of transformed models, each transformed model including a plurality of transformed model pixels. A hash table is constructed, each entry of the hash table corresponding to an image pixel that in turn corresponds to a transformed model pixel of at least one of the transformed models, each entry of the hash table including a list of all the transformed models that include that transformed model pixel. The feature pixels of the image are determined, and are augmented by a distance transformation to provide a set of relevant pixels. Each relevant pixel that appears in the hash table contributes a score to the corresponding transformed models. For each transformed model, the sums of the scores provides a tally.

Abstract: The present invention provides a method and apparatus for classifying image data. A video average may be determined from the image data. Peak detection may then be limited to an area within the image data where the video average is above a first threshold. Valley detection may be limited to an area where the video average is below a second threshold.

Abstract: The read out image data is stored in the mask memory in a predetermined mask unit, and binarized into binary value with a predetermined threshold in the binarizing circuit. The binarized image data is output to the main scanning direction-black pixel counting circuit and the sub scanning direction-black pixel counting circuit, respectively, to count black pixels in the main scanning direction and in the sub scanning direction, respectively. The largest value is determined by the comparator, and designated as the linkage degree. With the judging circuit, the area judgement of the character area, the photograph area and the dot area is performed based on the linkage degree.

Abstract: An apparatus including a method of digital image processing which allows a user to remove narrow, elongated areas of image distortion from a digitized image. According to the method, the user selects a portion of an image to be edited and, once the portion has been identified, the image distortion within the portion is automatically removed. The procedure for removing the distortion includes the steps of delineating the distortion within the portion of the image selected and replacing the delineated area with a correction image which is created as a function of the attributes of the neighboring portions of the image.

Abstract: A printed or handwritten character image is recognized by training (301) a plurality of lookup tables with a set of known referent characters; obtaining (302) a bitmap of a character image to be recognized; mapping (303) the bitmap onto a standardized character matrix; partitioning (304) the matrix into a plurality of sub-matrices; (305) determining a set of candidates for each sub-matrix; and selecting (306) a preferred candidate from among the set of candidates responsive to at least one pre-defined selection criterion. The invention is implemented by means of a scanner (102), a character mapper (104), a matrix partitioner (106), a candidate set builder (108), and a character selector (110).

Abstract: A method of processing digital image information, including pixel intensity information, includes the steps of identifying a region of the image in which pixel intensities are similar to each other, but the median pixel intensity differs significantly from the median pixel intensity of other parts of the image; and modifying pixel intensities in the identified region so that the median for the region is closer to the median for the other parts of the image. A related method of processing digital image information involves modifying pixel intensities according to a non-linear algorithm, with the result that one or more regions of the image have their median pixel intensities adjusted to a level which is closer to the median pixel intensity for other parts of the image.

Abstract: A system for identifying and tagging anomalies, such as images of dust and scratches, in a digital image so that they can be removed from the image. The invention implements a technique for detecting anomalies designed to minimize the number of false positives, i.e., non-anomalous portions of the image falsely identified as anomalies. The technique bases the initial first-pass identification of anomalous image regions on the difference between the gradient of the image at each of a set of grid points in the image and the mean of the gradient of the image at nearby points. Thus a region is identified as anomalous if original image pixel values in its immediate neighborhood are more variable than they are in a larger neighborhood. This technique greatly reduces (when compared to the prior art) the identification of false positives in noisy regions of the image, such as the leaves of a tree in daylight or a pebbly beach.

Abstract: The present invention provides a novel system and method that permits a “mask” to be directly incorporated into an image during image processing. This is accomplished by processing binary images or image data which are encoded using two bits rather than the usual one. The second bit is defined to be a “mask enable”, which directs a processor to pass the original data through to the output image regardless of the processing result for that pixel. The present invention also provides a means of automatically providing background data to the processor for pixels outside the original image so that the result image is always the same size as the original image. For binary images, the background may be defined to have a value of “0” or “1”, and this value is provided to the processing engine in place of all of the pixels which lie outside the original image.

Abstract: A gradient is calculated at each pixel of the image data to be filtered using a one-dimensional horizontal gradient operator of 1×2 size and a one-dimensional vertical gradient operator of 2×1 size, each of the gradient operators having a weighted value (1,−1). The absolute value of the calculated gradient is then compared with a predetermined threshold value to generate a binary value. This gradient calculation and comparison with a threshold is repeated for pixel blocks having a predetermined size to generate a binary edge map with respect to a whole block. A filter window of a predetermined size is then applied to the generated binary edge map to determine whether edges are present in the filter window. If no edges are present in the filter window, then the pixel values of the corresponding filter window pixel are filtered pixel-by-pixel by using a predetermined first set of weighted values to generate a new pixel value.

Abstract: A pattern inspection method capable of providing an improved inspection reliability includes the following steps. A pattern classification is set every pixel-value range of a reference-image data obtained from a reference pattern. Each of the pattern classifications has reference-data preparing parameters. With respect to each pixel of the reference-image data, pixel values of a pixel window which is composed of the pixel as a center pixel and neighbor pixels around the center pixel are checked to prepare a reference data. When all of the pixels in the pixel window are within a single pixel-value range, the center pixel is decided as a pattern uniform portion, and the reference data is prepared according to the parameters of a corresponding pattern classification.

Abstract: An image processing apparatus is provided with a linear processor array which includes a plurality of linearly coupled processing elements. The apparatus further includes a plurality of local memories each of which takes a form of one-dimensional pixel data memories. The local memories are respectively assigned to the plurality of processing elements. Further, the local memories are arranged so as to store in combination a two-dimensional pixel image to be processed. A plurality of buffers are respectively assigned to the processing elements. Each buffer stores pixel location information of at least one center pixel from which a unique pixel value is to be propagated if the center pixel exists on a corresponding local memory. With the arrangement mentioned above, in order to accelerate pixel data propagation in the two-dimensional pixel image, a plurality of register units are respectively assigned to the plurality of processing elements.

Abstract: Automatic image enhancement of halftone images is performed by subjecting the halftone image to a low-pass filter so as to smooth out halftone variations between adjacent pixels prior to performing conventional image analysis and image processing.

Abstract: A digitized image is treated by an electronic system where pairs of pixels within a convolution window are compared, and for those differences which are greater than a preselected, or automatically calculated, threshold, a black or white vector is counted, respectively, depending on whether the more centrally located pixel is darker or lighter than the outer pixel. The central pixel is replaced with an enhanced value, if it is surrounded by a majority of significantly lighter or darker pixels, as indicated by the black and white vector counts. Weighting factors allows for custom tailoring the enhancement algorithm to suit the need of the particular image.

Abstract: A one-dimensional signal adaptive filtering method and a one-dimensional signal adaptive filter.

Abstract: An image processing circuit of the present invention includes a succession calculating circuit. In the succession calculating circuit, an average value calculating circuit determines an average value of image data, inputted via line memories, in a specific region, and a multivalue-binary converting circuit converts, in accordance with the average value, the image data into binary data in which the entire image data in the specific region are separated into black pixels and white pixels. A succeeding black pixels calculating circuit calculates the number of succeeding black pixels each in a main scanning direction and in a sub scanning direction, and determines respective maximum values in the main scanning direction and in the sub scanning direction so as to output the respective maximum values as a main scanning direction maximum succession and a sub scanning direction maximum succession.

Abstract: An image processing method for detection and removal of halftone dots includes converting a gray scale image into a binary thresholded image with halftone dots; identifying halftone regions within the binary thresholded image; and removing halftone dots from the identified halftone regions. The identifying halftone regions step is effected by classifying the binary thresholded image to produce a halftone classification map; reclassifying the halftone classification map to produce a halftone reclassification map of lines with plural halftone pixels in each line; merging the halftone pixels in each line in the halftone reclassification map to produce a halftone line map; and merging the lines of the halftone line map to produce a halftone region map.

Abstract: An edge detection method employs binary morphological erosion. A binary image is generated from the gray-scale-value input image. A structure element is guided in a step-by-step manner across the binary image and generates an eroded binary image in accordance with an erosion rule. By forming the difference between the binary image and the eroded binary image, an output image containing the edges is generated. An output image which contains masked edges is generated through the use of a further erosion rule. The further erosion rule is based on a gray-scale value threshold and is applied to the eroded binary image to form a twice-eroded binary image. The difference between the twice-eroded binary image and the binary image forms the image which contains masked edges.

Abstract: A multiresolutional filter called a critical point filter is introduced. This filter extracts a maximum, a minimum, and two types of saddle points of pixel intensity for every 2.times.2 (horizontal.times.vertical) pixels so that an image of a lower level of resolution is newly generated for every type of a critical point. Using this multiresolutional filter, a source image and a destination image are hierarchized, and source hierarchical images and destination hierarchical images are matched using image characteristics recognized through a filtering operation.

Abstract: An approach to synthesizing missing pixels in video images. The invention synthesizes an outermost ring of pixels within a block of missing pixels, based on adjacent, known, pixels. Then, the invention synthesizes the next-outermost ring, using, in part, pixels synthesized in the outermost ring, and proceeds in this manner to the innermost ring.

Abstract: A process for reducing background clutter of a scene, and isolating an object of interest. The process includes calculating a minimum output from a plurality of appropriately configured anti-median filters representing a minimum difference processor (MDP) filter. In a first embodiment, the MDP filter includes a (5.times.5) matrix of elements wherein each element has an intensity value representation of the scene. A first anti-median filter array is configured as a horizontal array of 5 pixel elements in which the center element of the matrix is the center element of the horizontal array. A median value of the 5 elements of the horizontal array is determined, and is subtracted from the value of the center element to determine an anti-median value of the horizontal array. The same process is performed for a vertical 5 element anti-median array, and two diagonal 5 element anti-median arrays, all including the center element of the matrix as the center common element of the arrays.

Abstract: A method and apparatus for detecting objects in computed tomography (CT) data, including sheet-shaped objects such as sheet explosives can be detected by analyzing a neighborhood of voxels surrounding a test voxel. If the density of the test voxel is sufficiently different from the mean density of the neighboring voxels, then it is concluded that the test voxel is associated with a sheet object. A connectivity process can be applied to voxels to combine them into objects after sheets are detected to prevent sheets from being inadvertently removed from the data by erosion.

Abstract: A system and method for reducing noise using recursive noise level estimation. The system and method for noise reduction substitute a target pixel in a processing window with a weighted average of a plurality of neighboring pixels according to the degree of similarity between thc target pixel and the neighboring pixels. The similarity is based on the noise level affecting the image and the local brightness of the processing window. The filter is based on fuzzy logic and filters out noise without smoothing the image's fine details. The filter uses a human visual system (HVS) response to adjust brightness.

Abstract: A method and apparatus for editing continuous tone digital images representing pixels in which regions containing defects, called "defect regions", are unnoticeable or less visually discernible. Corrected pixel data for pixels in the defect region, called defect pixels, in the digital image are provided while data representing non-defect pixels outside of the defect region are not altered. These digital images may have one color channel (gray side) or multiple color channels. The apparatus and method includes a computer system which operates according to a set of instructions.

Abstract: In the image processing apparatus of the present invention, the dot pattern analysis section determines a plurality of black pixels neighboring a notice pixel of a predetermined positioned black pixel in a window of the binary image and determines a white pixel as a compensation pixel in the window in accordance with coordinates of the notice pixel and the plurality of black pixels. A notice pixel multivalued section converts the black pixel of the notice pixel to a multivalued half tone pixel and outputs the multivalued half tone pixel based on the coordinate of the notice pixel. A compensation pixel multivalued section converts the white pixel of the compensation pixel to a multivalued half tone pixel and outputs the multivalued half tone pixel based on the coordinate of the compensation pixel.

Abstract: A denomination recognition apparatus and method uses a programmed microelectronic CPU (14) to execute program instructions (19) stored in PROM (18) to read in pixel data from an optical imaging section including LEDs (15) and photodiodes (16) for generating signals which can be converted to a first image of a currency note (13) being transported along the path of travel. The CPU (14) receives position and skew data detected by external sensors for sensing the position and skew of the note (13).

Abstract: A method and apparatus for detecting objects in computed tomography (CT) data, including sheet-shaped objects such as sheet explosives, analyze a neighborhood of voxels surrounding a test voxel. Erosion of the data can be performed by identifying a neighborhood of voxels surrounding a voxel of interest. If the number of voxels having densities below a predetermined threshold exceeds a predetermined number, then it is assumed that the test voxel is a surface voxel and is removed from the object. A connectivity process can be applied to voxels to combine them into objects after sheets are detected to prevent sheets from being inadvertently removed from the data by erosion. A dilation function can then be performed on the eroded object to replace surface voxels removed by erosion.

Abstract: A method and apparatus for detecting objects in computed tomography (CT) data are disclosed. Erosion of the data can be performed by identifying a neighborhood of voxels surrounding a voxel of interest. If the number of voxels having densities below a predetermined threshold exceeds a predetermined number, then it is assumed that the voxel of interest is a surface voxel of an object and is removed from the object. A connectivity process is then applied to remaining voxels to combine them into objects. A dilation function can then be performed on the eroded object to replace surface voxels removed by erosion. This morphological connected components labeling (CCL) approach separates adjacent objects in the data such that they can be individually labeled and analyzed.

Abstract: A degree of frequency and a comparison difference value are used as an amount of feature that is to be inputted to a discrimination circuit so as to discriminate the kind of regions. The degree of frequency is represented by the smaller of two sums, that is, the sum of absolute values of the density-level difference between pixels adjoining in the line direction of the pixels within the specific region and the sum of absolute values of the density-level difference between pixels adjoining in the direction orthogonal to the line of the pixels, which is obtained by a frequency-degree extracting circuit. Further, the comparison difference value is represented by the comparison difference between a maximum value and a minimum value in the density levels of all pixels within the specified region, which is obtained by the comparison difference circuit.

Abstract: In the image output apparatus of the present invention, a window forming section extracts a window including a notice pixel from bit map data of the image. A contour extraction section extracts contour pixels from the window based on the position of the notice pixel. An outline estimation section estimates an outline based on positions of the contour pixels. A smoothing signal generation section generates a smoothing signal for the notice pixel based on the relative position of the notice pixel and the outline.

Abstract: An apparatus for and a method of reliably detecting a very small dotted image area. A detecting area including a predetermined number of pixels including a target pixel is set. It is judged whether or not the detecting area belongs to a dotted image area on the basis of the number of edge pixels in the detecting area. Consequently, it is possible to detect edges peculiar to the dotted image area on the basis of a leading edge portion or a trailing edge portion of the density, to detect the very small dotted image area on the basis of the number of edge pixels.

Abstract: A method of processing a digital image representing ribbon-shaped objects of non-uniform intensity contrasting with a background of lower intensity includes an automatic segmentation phase having one or more morphological opening operations effected, respectively, with one or more three-dimensional structuring elements. The latter have a two-dimensional base parallel to the image plane and have a non-binary intensity function in a third dimension. Preferably, the automatic segmentation phase is carried out by means of a set of two-dimensional spatial structuring elements with a third intensity dimension. The set contains N anisotropic structuring elements oriented from .pi./N to .pi./N and one isotropic structuring element.

Abstract: A process for detecting defects in a textured suface, particularly on a highway. The process consists in displacing with respect to the surface a camera presenting an inclination and in processing by processing systems the successive images taken by the camera. A mathematical processing is effected on sub-images, corresponding to the same spatial portion of the surface, taken at different consecutive instants.

Abstract: A multiresolutional filter called a critical point filter is introduced. This filter extracts a maximum, a minimum, and two types of saddle points of pixel intensity for every 2.times.2 (horizontal x vertical) pixels so that an image of a lower level of resolution is newly generated for every type of a critical point. Using this multiresolutional filter, a source image and a destination image are hierarchized, and source hierarchical images and destination hierarchical images are matched using image characteristics recognized through a filtering operation.

Abstract: Apparatus and a method for semi-automatic retouching of digital images is provided. The method includes the steps of processing the digital image to obtain a list of candidate defect segments, computing statistics for a list of candidate defect segments and for the segments of the image surrounding the candidate defect segments, selecting a sublist of defect segments in accordance with pre-determined criteria using the computed statistics, displaying the image together with the sublist and amending the values of the sublist of defect segments to obtain a retouched image.

Abstract: A method and apparatus for resizing or processing images initially retrieves a one-dimensional sample of the image, such as a line of pixels. The image can be of a machine-readable symbol, or other data. A moving window having, for example, five pixels, includes a center "hard core" and four surrounding pixels. Two or more structural sets or multisets are applied to the window to produce a set of pixels greater than the set of pixels in the original image. For example, the center pixel is repeated three times, while immediately adjacent left and right pixels are repeated twice. Thus, a series of nine pixels results. The resulting nine pixels are then sorted based on descending gray scale levels. The n-th largest pixels is selected, such as the third largest. The image is updated, the window moved, and the method repeats again. Pyramid filtering, cubic convolution interpolation, or other techniques can be used to reduce the size of an image, and more complex windows and structural sets can be employed.

Abstract: A method of identifying defective pixels included in a digital image. The method includes the steps of designating a pixel adjacent the defective pixels in the digital image, setting an evaluation area, setting a defective pixel candidate, computing, as a line average, an average pixel value of each of lines above, below, leftward and rightward of the defective pixel candidate, diminishing the evaluation area by selecting an upper line, a lower line, a left line and a right line, computing a threshold pixel value for determining the defective pixels, from an average pixel value of each line, and comparing values of all pixels with the threshold pixel value to determine the defective pixels.

Abstract: Upon measuring the image-quality of an original document, the original document is photographed, and the resulting image is displayed on cells within a window so that image-quality data, such as the degree of edge-irregularity a, is obtained. A regression line is found based upon the correlation between the cell size and the image-quality data. By setting the image-quality data resulting from a specific sell size as a reference value, respective pieces of image-quality data are converted and corrected based upon the regression line using the reference value as a target. Thus, it becomes possible to compare pieces of image quality data in an equivalent manner even if they were obtained from measurements using different magnifications.

Abstract: An apparatus polygonally approximates a contour of an object expressed in a digital video signal. First of all, the apparatus determines a pair of vertices on the contour. Then, a line segment connecting the pair of vertices is generated and widened to thereby produce a band segment. And, a contour segment corresponding to the line segment is detected and, subsequently, the band segment and the contour segment are matched. The band segment is widened to determine a new vertex located between the pair of vertices on the contour until the contour segment is completely covered by the widened band segment. During the band segment widening process, at least one contour pixel on the contour segment, wherein the detected contour pixels, if there are more than one, are last pixels being covered by the widened band segment, are determined and, among the detected contour pixels, a contour pixel, which is closest to a center of a straight line joining the pair of vertices, is determined as the new vertex.

Abstract: A loop filtering method and a loop filter therefor are provided.

Abstract: An automatic recognition method capable of accurately determining edges of a concavity or convexity in order to achieve an improved measuring accuracy in recognition of a concavity or convexity. This method comprises the steps of: converting a photographic image of an object having a concavity or convexity defined therein into data in a specified feature space; obtaining a plurality of points that exist outside a virtual circle determined by a virtual center of the concavity or convexity, the virtual center being established for each elementary area of the image converted into the feature space data; obtaining the equation of a plane that passes through the plurality of points; detecting symmetric pairs of pixels in each elementary area of the feature space data with respect to the corresponding virtual concavity or convexity center; and recognizing the position of the concavity or convexity of the object when the symmetric pixels satisfy the equation of the plane.

Abstract: A window texture extraction process for quantifying texture features in grey-level images and combining the quantified texture features with previously derived classification results. The previously determined classification results require that all of the training data and all the target data be assigned, or be capable of being assigned to one of two classes. The result is a set of new features that can be quickly analyzed to determine such properties as class membership for identification purposes. In order to fold-in the classification results, the class membership is established in previous experiments on images which have been characterized. The previous experiments are executed through the application of a window texture extraction kernel according to the invention.

Abstract: A convolver includes a plurality of multipliers for multiplying pixel values of a convolution window by corresponding coefficients of a convolution mask to provide products, a summer for summing the products to provide a result and a memory for storing intermediate results. The convolver may be used to perform an N.times.N convolution in two or more passes. A first subset of pixel values of an N.times.N convolution window and a first subset of corresponding coefficients of an N.times.N convolution mask are supplied to the multipliers during a first pass of the N.times.N convolution. The summer provides an intermediate result for the first pass and stores the intermediate result in the memory. A second subset of pixel values of the N.times.N convolution window and a second subset of corresponding coefficients of the N.times.N convolution mask are supplied to the multipliers during a second pass of the N.times.N convolution.

Abstract: The ringing detector of the invention receives data representing pixel values of a plurality of pixels including: a remarked pixel; a number L (where L is a natural number) of neighboring pixels which are located in the vicinity of the remarked pixel; and a number M (where M is a natural number) of surrounding pixels which are more distant from the remarked pixel than the neighboring pixels are, thereby generating a ringing constant t indicating whether or not a ringing exists in the vicinity of the remarked pixel. The ringing detector includes: a pixel value variation detector for calculating a value x corresponding to a variation in the pixel values of the neighboring pixels; a distance detector for calculating a value y corresponding to a distance between the pixel values of the surrounding pixels and a barycenter of the pixel values of the remarked pixel and the neighboring pixels; and a ringing constant calculator for calculating the ringing constant t based on the values x and y.

Abstract: In a dynamic image processing apparatus of the present invention, the focus of contraction of the optical flow of the background in the dynamic image is extracted. When a series of dynamic images are inputted, an optical flow extraction section extracts an optical flow representing motion vector of each block between two neighboring dynamic images. A Hough transform section votes each optical flow of the dynamic image to vote space by unit of block. The vote space comprises two planes and one circle positioned in parallel to each other. The dynamic image is included in one plane. The other plane is positioned symmetrically to the one plane with respect to a center position of the circle. The Hough transform section selects a block whose voted value is maximum in all blocks of the vote space. The selected block corresponds to the focus of contraction of the optical flow of background in the dynamic image.

Abstract: Light is obliquely irradiated onto the surface of a document in braille, and its reflected light is received, to obtain a first gray image. The first gray image is subjected, while being scanned in a direction corresponding to a direction parallel to the direction of light irradiation, to differentiation processing in the scanning direction, to produce a second gray image whose light and dark portions respectively produced by projected points have a particular pattern. The second gray image is subjected to such ternary-coding processing that the particular pattern of the light and dark portions respectively produced by the projected points in the second gray image is emphasized, to produce a ternary-coded image. Each of the projected points is extracted by pattern matching with respect to the ternary-coded image.

Abstract: A technique combines most of the advantages of both blue noise and clustering in generating a clustered aperiodic mask and using it in a dither array algorithm for halftoning. The method of halftoning of grey scale images utilizes a pixel-by-pixel comparison of the image against a clustered aperiodic mask in which the clustered aperiodic mask is comprised of a partly random and partly deterministic single valued function which is designed to produce visually pleasing dot configurations when thresholded at any level of grey.

Abstract: A method to extract automatic high level features of a gray level ridge flow image including the automatic determination of the location of the high level features including the core and delta points of a fingerprint image in the form of a gray level image, comprising the following five steps:(1) determining a ridge flow direction of each gray level pixel of a fingerprint image and assign it with a direction code;(2) finding out a block directional flow and assigning each with appropriate direction code thereof;(3) correcting block flow directions in which a detection of the flow direction is affected image an input image quality in order to be immune from noises infiltrated during the acquiring process;(4) locating the delta points which are based on a corrected block direction flow diagram to locate from zero to a few delta points; and(5) locating the core points which are based on a corrected block direction flow diagram to locate one to two core points.

Abstract: An image processing apparatus includes a converter for reading out a manuscript image signal and converting the read-out signal to an electric signal. A signal processor executes signal processing for the electric signal, and an image generator generates an image in accordance with the processed signal. An outputting unit separates a character area and a photograph area of the manuscript from the read-out signal and outputs an area signal. A changing-over circuit changes over a filtering process in accordance with the area signal. A halftone processor executes in parallel a halftone process for the character area image and a halftone process for the photograph area image, and the changing-over circuit selectively changes over the halftone processing result for the character and the halftone processing result in accordance with the area signal.

Abstract: A method of removing columnar streaks from a digital image of the type in which it is assumed that pixels in a predetermined region near a given pixel are strongly related to each other and employing gain and offset values to compute streak removal information, a test is performed for a strong relation between the pixels in a predetermined region near a given pixel and streak removal information is computed only if such a strong relationship exists, whereby image content that does not extend the full length of the image in the column direction will not be interpreted as a streak.

Abstract: A method for determining an orientation of a line segment in a contour in a local area of a binary contour image. The number of pixels having a predetermined value and located in each orientation in the local area, is obtained, and it is determined that a line segment exists in an orientation in which the number of pixels is large. Further, when it is determined that the number of pixels located in each of a plurality of orientations and having a predetermined value, is greater than a sum of the numbers of pixels in orientations adjacent to said each orientation on both sides thereof, it is determined that a line segment exists in said each orientation. Two orientations in which large numbers of pixels are located are detected, and it is determined whether the point is a constituent of a line segment, a line segment of the sub-straight-line form, or a near-corner point, depending on angles made by the two orientations is equal to 180.degree.