System and method for image processing using segmentation of images and classification and merging of image segments using a cost function

Claim 2

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component fails, wherein at least one of said classes indicating that the connected component is most likely to be a single character, at least one of the classes indicating that the connected component is most likely to be less than a single character, and at least one of said classes indicating that the connected component is most likely to be more than a single character; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components progressively with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components.

Claim 3

Apparatus as claimed in claim 1 wherein the classification logic comprises means to derive a feature vector from the connected components via a two-pass process in which features of a document image in which the image is located are used to normalize the feature values.

Claim 4

Apparatus as claimed in claim 1 wherein the splitting logic is arranged to operate before the merging logic.

Claim 5

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component fails, at least one of said classes indicating that the connected component is most likely to be a single character, wherein the classification logic comprises a neural network; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components, progressively, with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components.

Claim 6

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component fails, at least one of said classes indicating that the connected component is most likely to be a single character, wherein the classification logic comprises a neural network; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components, progressively, with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components.

Claim 7

Apparatus as claimed in claim 4 wherein the classification logic comprises means to derive a feature vector from the connected components via a two-pass process in which features of a document image in which the image is located are used to normalize the feature values.

Claim 8

Apparatus as claimed in claim 4 wherein the splitting logic is arranged to operate before the merging logic.

Claim 9

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character, wherein the classification logic has a tree structure comprising partial classifiers at each node; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components progressively with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value far each of the possible remaining pairs of connected components.

Claim 10

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character, wherein the classification logic has a tree structure comprising partial classifiers at each node; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components progressively with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value far each of the possible remaining pairs of connected components.

Claim 11

Apparatus as claimed in claim 7 wherein the classification logic comprises means to derive a feature vector from the connected components via a two-pass process in which features of a document image in which the image is located are used to normalize the feature values.

Claim 12

Apparatus as claimed in claim 7 wherein the splitting logic is arranged to operate before the merging logic.

Claim 13

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the splitting logic is arranged to generate a tree structure by postulating a single split hypothesis for each connected component, the images generated by the split hypothesis being processed by the classification logic to accept or reject the split and if the split is accepted being added to the tree as children nodes and processed by the splitting logic in the same way until no further splitting is possible, and wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components progressively with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components.

Claim 14

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the splitting logic is arranged to generate a tree structure by postulating a single split hypothesis for each connected component, the images generated by the split hypothesis being processed by the classification logic to accept or reject the split and if the split is accepted being added to the tree as children nodes and processed by the splitting logic in the same way until no further splitting is possible, and wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to calculate a combining cost value for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components progressively with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components.

Claim 15

Apparatus as claimed in claim 10 wherein the classification logic comprises means to derive a feature vector from the connected components via a two-pass process in which features of a document image in which the image is located are used to normalize the feature values.

Claim 16

Apparatus as claimed in claim 10 wherein the splitting logic is arranged to operate before the merging logic.

Claim 17

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to generate a matrix of combining cost for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components, progressively, with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, the images generated by the combination being processed by the classification logic to accept or reject the combination until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components, and to repeat the process until a stopping criterion is satisfied.

Claim 18

Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising: identification logic for identifying connected components in the input image; classification logic for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character; and merging logic and splitting logic for merging and splitting the connected components, respectively, the merging and splitting logic and the classification logic being arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components are reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion, wherein the merging logic performs the merging of the connected components by first evaluating a pre-determined cost function to generate a matrix of combining cost for each possible pair of connected components in a field, and then attempting to merge the pairs of connected components, progressively, with increasing combining cost values, commencing with the pair of connected components having a minimum combining cost value, the images generated by the combination being processed by the classification logic to accept or reject the combination until one of the attempted merges is accepted by the classification logic; the merging process being continuously repeated after each acceptance of an attempted merge by the classification logic by evaluating the predetermined cost function to calculate a combining cost value for each of the possible remaining pairs of connected components, and to repeat the process until a stopping criterion is satisfied.

Claim 19

Apparatus as claimed in claim 13 wherein the classification logic comprises means to derive a feature vector from the connected components via a two-pass process in which features of a document image in which the image is located are used to normalize the feature values.

Claim 20

Apparatus as claimed in claim 13 wherein the splitting logic is arranged to operate before the merging logic.