Image recognition method and apparatus for the same method

An image recognition apparatus comprising:

• an image pickup device configured to take a first image;
• an image-element divider configured to extract a plurality of elements having different shapes constructing an entire character included in said first image to obtain a second image, in which each of said elements is enclosed by an inner rectangular frame with a shape individually correlated to a shape of the element surrounded by the inner rectangular frame;
• a composite-image generator for forming a composite image with respect to a combination of said inner rectangular frames in said second image;
• a characteristic-amount calculator for determining a characteristic amount of said elements included in said composite image;
• a back-propagation network, in which learning about a reference character(s) to be included in said first image has already finished, for providing a degree of agreement between the characteristic amount of said composite image and the reference character when the characteristic amount is input into said back-propagation network;
• and
• an image analyzer for determining said composite image having a highest degree of agreement between the characteristic amount of said composite image and the reference character from results provided by said back-propagation network with respect to different combinations of said inner rectangular frames in said second image, to output it as a recognition data.

An image recognition method comprising:

• performing the following steps on at least one processor
• (I) taking a first image including a character composed of plural elements;
• (II) extracting said plural elements in said first image to obtain a second image, in which each of said plural elements is enclosed by an inner rectangular frame;
• (III) forming a composite image with respect to a combination of said inner rectangular frames in said second image;
• (IV) calculating an area of said elements included in said composite image;
• (V) inputting said area in a back-propagation network, in which learning about a reference character(s) to be included in said first image has already finished, to provide a degree of agreement between the area of said composite image and the reference character;
• and
• (VI) determining said composite image having a highest degree of agreement between the area of said composite image and the reference character from results obtained by repeating the steps (III) to (V) with respect to different combinations of said inner rectangular frames in said second image, to output it as a recognition data;
• the method further comprising
• enclosing at least two of the inner rectangular frames with an outer rectangular frame;
• comparing a calculated area of the outer rectangular frame with a required value previously stored in the back propagation network;
• and
• adding at least a third inner rectangular frame to the outer rectangular frame if the calculated area of the outer rectangular frame is less than the required value.

An image recognition method comprising:

• performing the following steps on at least one processor
• (I) taking a first image including a character composed of plural elements;
• (II) extracting said plural elements in said first image to obtain a second image, in which each of said plural elements is enclosed by an inner rectangular frame;
• (III) forming a composite image with respect to a combination of said inner rectangular frames in said second image;
• (IV) calculating a characteristic amount of said elements included in said composite image;
• (V) inputting said characteristic amount in a back-propagation network, in which learning about a reference character(s) to be included in said first image has already finished, to provide a degree of agreement between the characteristic amount of said composite image and the reference character;
• and
• (VI) determining said composite image having a highest degree of agreement between the characteristic amount of said composite image and the reference character from results obtained by repeating the steps (III) to (V) with respect to different combinations of said inner rectangular frames in said second image, to output it as a recognition data,
• wherein the combination of said inner rectangular frames in said second image is defined by surrounding at least two inner rectangular frames with an outer rectangular frame,
• a calculated area of said outer inner rectangular frame is compared with a required value previously stored in the back propagation network,
• at least a third inner rectangular frame is added to said outer rectangular frame to define another outer rectangular frame when the calculated area of said outer rectangular frame is less than the required value, and
• said another outer rectangular frame is stored as a candidate image for said composite image when said another outer rectangular frame is equal to or larger than the required value.

An image recognition apparatus comprising:

• an image pickup device configured to take a first image;
• an image-element divider configured to extract a plurality of elements constructing a character included in said first image to obtain a second image, in which each of said elements is enclosed by an inner rectangular frame;
• a composite-image generator configured to form a composite image with respect to a combination of said inner rectangular frames in said second image;
• a characteristic-amount calculator configured to determine a characteristic amount of said elements included in said composite image;
• a back-propagation network, in which learning about a reference character(s) to be included in said first image has already finished, for providing a degree of agreement between the characteristic amount of said composite image and the reference character when the characteristic amount is input into said back-propagation network;
• and
• an image analyzer configured to determine said composite image having a highest degree of agreement between the characteristic amount of said composite image and the reference character from results provided by said back-propagation network with respect to different combinations of said inner rectangular inner frames in said second image, to output it as a recognition data,
• wherein said composite-image generator defines the combination of said rectangular frames in said second image by surrounding at least two rectangular frames with an outer rectangular frame,
• a calculated area of said outer rectangular frame is compared with a required value previously stored in the back propagation network,
• said composite-image generator defines another outer rectangular frame by adding at least a third rectangular frame to said outer rectangular frame when the calculated area of said outer rectangular frame is less than the required value, and
• said another outer rectangular frame is stored as a candidate image for said composite image in a memory when said another outer rectangular frame is equal to or larger than the required value.