초록 ▼

A method for recognizing an object image comprises the steps of extracting a candidate for a predetermined object image from an overall image, and making a judgment as to whether the extracted candidate for the predetermined object image is or is not the predetermined object image. The candidate for

A method for recognizing an object image comprises the steps of extracting a candidate for a predetermined object image from an overall image, and making a judgment as to whether the extracted candidate for the predetermined object image is or is not the predetermined object image. The candidate for the predetermined object image is extracted by causing the center point of a view window, which has a predetermined size, to travel to the position of the candidate for the predetermined object image, and determining an extraction area in accordance with the size and/or the shape of the candidate for the predetermined object image, the center point of the view window being taken as a reference during the determination of the extraction area. A learning method for a neural network comprises the steps of extracting a target object image, for which learning operations are to be carried out, from an image, feeding a signal, which represents the extracted target object image, into a neural network, and carrying out the learning operations of the neural network in accordance with the input target object image.

대표청구항 ▼

A method for recognizing an object image comprises the steps of extracting a candidate for a predetermined object image from an overall image, and making a judgment as to whether the extracted candidate for the predetermined object image is or is not the predetermined object image. The candidate for

A method for recognizing an object image comprises the steps of extracting a candidate for a predetermined object image from an overall image, and making a judgment as to whether the extracted candidate for the predetermined object image is or is not the predetermined object image. The candidate for the predetermined object image is extracted by causing the center point of a view window, which has a predetermined size, to travel to the position of the candidate for the predetermined object image, and determining an extraction area in accordance with the size and/or the shape of the candidate for the predetermined object image, the center point of the view window being taken as a reference during the determination of the extraction area. A learning method for a neural network comprises the steps of extracting a target object image, for which learning operations are to be carried out, from an image, feeding a signal, which represents the extracted target object image, into a neural network, and carrying out the learning operations of the neural network in accordance with the input target object image. he characters in each of the candidate regions to each of characters in each of predetermined words in order to determine how many characters coincide; andcalculating a ratio of a denominator which is the larger of the number of characters of each of the candidate regions and the number of characters of each predetermined word, to a numerator which is the number of the coincided characters, and, in case that the calculated ratio is larger than the determination probability, the candidate region is determined to be a word region.4. The method as recited in claim 3, wherein the step of determining the abstract content portion includes the steps of:selecting a region in which a vertical direction start coordinate is equal to or larger than that of a predetermined region corresponding to the abstract region and is smaller than that of a predetermined region corresponding to the introduction region, when two words representing the abstract and the introduction exist and the abstract content portion is made by one column;obtaining a minimum value and a maximum value, when a vertical direction last coordinate of the selected region is larger than that of a comparing region, the minimum value being replaced with the vertical direction start coordinate of the selected region and the maximum value being replaced with the vertical direction last coordinate of the comparing region, and when the vertical direction last coordinate of the selected region is less than that of the comparing region, the maximum value being replaced with the vertical direction start coordinate of the selected region and the minimum value being replaced with the vertical direction last coordinate of the comparing region; anddetermining whether the selected region is the abstract region with the abstract content portion or not by using both the maximum and the minimum values.5. The method as recited in claim 3, wherein the step of determining the abstract content portion includes the steps of:selecting a region in which a horizontal direction last coordinate is located left of a center of the document image, when two words representing the abstract and the introduction exist and the abstract content portion is positioned at a left column of two columns in the document image; andselecting a region, among the selected regions, in which a vertical direction start coordinate is equal to or larger than that of a predetermined region corresponding to the abstract region and smaller than that of another predetermined region corresponding to the introduction region.6. The method as recited in claim 3, wherein the step of determining the abstract content portion includes the steps of:selecting a region in which a horizontal direction last coordinate is located left of a center of the document image and a vertical direction start coordinate is equal to or larger than that of a predetermined region corresponding to the abstract region, when two words representing the abstract and the introduction exist and the abstract content portion is positioned in both left and right columns in the document image; andselecting a region, among the selected regions, in which a horizontal direction start coordinate is located right of the center of the document image and a vertical direction last coordinate is smaller than that of the region corresponding to the introduction.7. The method as recited in claim 3, wherein the step of determining the abstract content portion includes the steps of:selecting a region in which a vertical direction start coordinate is equal to or larger than that of a predetermined region corresponding to the abstract region, when a word representing the abstract exists and a word representing the introduction does not exist and the abstract content portion is made by one column in the document image; obtaining a minimum value and a maximum value, when a vertical direction last coordinate of the selected region is larger than that of a comparing region, the minimum value being replaced with the vertical region start coordinate of the selected region and the maximum value being replaced with the vertical direction last coordinate of the comparing region, and when the vertical direction last coordinate of the selected region is less than that of the comparing region, the maximum value being replaced with the vertical direction start coordinate of the selected region and the minimum value being replaced with the vertical direction last coordinate of the comparing region; anddetermining whether the selected region is the abstract region with the abstract content portion or not by using both the maximum and the minimum values.8. The method as recited in claim 3, wherein the step of determining the abstract content portion includes the steps of:selecting a region in which a horizontal direction last coordinate is located left of a center of the image document and a vertical direction start coordinate is equal to or larger than that of a region corresponding to the abstract region, when a word representing the introduction does not exist and a word representing the abstract exists and the abstract content portion is positioned at two columns in the document image; andselecting a region in which a horizontal direction start coordinate is located right of the center of the document image.9. The method as recited in claim 1, wherein the title and the author are separated using the number of regions, and, when the number of regions containing the title and author is two, a first region is set to the title and a second region is set to the author, and, when the number of regions containing the title and author is four, a font size is compared to set the first and a second regions to the title and a third region and a fourth region to the author, and, when the title and author are separated by the basic structure form, it is determined whether the number of the regions is equal to a number of items of the type definition and whether a region arrangement structure is identical to an item arrangement structure.10. The method as recited in claim 9, wherein the step of separating the title and author includes the steps of:setting, when the number of the regions containing the title and author is two and the font size of characters of the second region is less than a predetermined value, the first region to the title and the second region to the author;setting, when the number of the regions containing the title and author is four and the sizes of the characters of the first and the second regions is larger than a predetermined size of the other regions, the first and the second regions to the title, and the third and the fourth regions to the author;setting, when the number of regions containing the title and the author is four and the font sizes of the characters of the first and the third regions are larger than a predetermined size of the other regions, the first and the third regions to the title and the second and the fourth regions to the author; andseparating, when the number of regions containing the title and author is not two or four, the title and the author by determining whether the number of the regions is equal to the number of items of the type definition and whether the region arrangement structure is identical to the item arrangement structure.11. A computer readable media containing a program for analyzing a structure of a document image to make a table of contents having a title, an author and an abstract, the program having functions of: dividing the document image into a number of regions and classifying the divided regions into text regions and non-text regions ac cording to attributes of the regions;selecting candidate regions representing an abstract region and an introduction region, and finding word regions from the candidate regions to determine a position of an abstract content portion;separating the title and the author using a basic form structure and a type definition representing an arrangement of journals, said candidate regions being selected from the divided regions when a horizontal length of each of the divided regions is smaller than eight times an average horizontal length of the text regions, and said word regions being extracted from the candidate regions based on a determination probability; andrecognizing a content of the title, author and abstract region to generate said table of contents. ; US-5480839, 19960100, Ezawa et al., 437/209; US-5481118, 19960100, Tew, 250/551; US-5481133, 19960100, Hsu, 257/621; U