Graph editing distance solving method for letter recognition

Abstract

The invention discloses a graph editing distance solving method for letter recognition. According to the method, based on an existing SFBP algorithm framework, the number of elements meeting a constraint condition in each row and each column is counted through item-by-item search comparison of the elements of a cost matrix framework; and by comparing the ratio of the number of the elements meetingthe constraint condition in each row and each column to the elements in each row and each column, a corresponding number of rows and columns of cost values is added to the SFBP cost matrix frameworkto change the cost matrix framework, and therefore the purpose of optimization is achieved. When the optimization objective is achieved, a solving algorithm can be used to perform solving calculationon a cost matrix, therefore, the limitation of the constraint condition on algorithm usage is avoided, and the algorithm is better applied to the letter recognition field.

Description

technical field

[0001] The invention relates to the technical field of graph edit distance, in particular to a graph edit distance method for letter recognition. Background technique

[0002] Recognition of letters is a common problem in pattern recognition. After the feature extraction is completed, the letters are mainly composed of points and lines. For the recognition speed, the faster the better, and the noise data may be generated after the feature extraction. Therefore, an algorithm with a faster processing speed and better fault tolerance is required. The algorithm based on graph edit distance is one of the common algorithms in the field of letter recognition.

[0003] The graph edit distance method is one of the most fault-tolerant and widely used methods in graph matching technology. The graph edit distance refers to the sum of the minimum editing operation costs required to edit the source graph into the target graph. There are six editing operations, namely ins...

Examples

Embodiment 1

[0041] A graph edit distance method for letter recognition, including the following steps:

[0042] The first step is to construct the cost matrix framework of SFBP algorithm.

[0043]Step 11. Use the relevant algorithm to extract the features of any two letters as graph data. Assume that the number of vertices of the two graph data obtained is a and b respectively, each vertex has a digital label, and the edge structure between vertices is undirected and unlabeled. For the convenience of description, according to the number of vertices in the graph, the graphs are called a graph and b graph respectively.

[0044] Step 12. Set the insert operation cost value C i and delete operation cost C d . Set insert operation cost value C i and delete operation cost C d Can be the same or different depending on the point label value.

[0045] Step 13: Use the Euclidean distance method to calculate the label distance of all vertices between the two graphs, and the obtained label dis...

Embodiment 2

[0063] The method of the second embodiment is substantially the same as that of the first embodiment, the difference is that, in the first embodiment, after calculating the number of cost values ​​that satisfy the constraints in each row and each column, these cost values ​​are replaced by positive infinity at the same time,

[0064] Another graph edit distance method for letter recognition includes the following steps:

[0065] Step 1. Construction of the cost matrix framework of SFBP algorithm.

[0066] Step 11. Use the relevant algorithm to extract the features of any two letters as graph data. Assume that the number of vertices of the two graph data obtained is a and b respectively, each vertex has a digital label, and the edge structure between vertices is undirected and unlabeled. For the convenience of description, according to the number of vertices in the graph, the graphs are called a graph and b graph respectively.

[0067] Step 12. Set the insert operation cost v...

technical field

[0001] The invention relates to the technical field of graph edit distance, in particular to a graph edit distance method for letter recognition. Background technique

[0002] Recognition of letters is a common problem in pattern recognition. After the feature extraction is completed, the letters are mainly composed of points and lines. For the recognition speed, the faster the better, and the noise data may be generated after the feature extraction. Therefore, an algorithm with a faster processing speed and better fault tolerance is required. The algorithm based on graph edit distance is one of the common algorithms in the field of letter recognition.

[0003] The graph edit distance method is one of the most fault-tolerant and widely used methods in graph matching technology. The graph edit distance refers to the sum of the minimum editing operation costs required to edit the source graph into the target graph. There are six editing operations, namely ins...

Examples

Embodiment 1

[0041] A graph edit distance method for letter recognition, including the following steps:

[0042] The first step is to construct the cost matrix framework of SFBP algorithm.

[0043]Step 11. Use the relevant algorithm to extract the features of any two letters as graph data. Assume that the number of vertices of the two graph data obtained is a and b respectively, each vertex has a digital label, and the edge structure between vertices is undirected and unlabeled. For the convenience of description, according to the number of vertices in the graph, the graphs are called a graph and b graph respectively.

[0044] Step 12. Set the insert operation cost value C i and delete operation cost C d . Set insert operation cost value C i and delete operation cost C d Can be the same or different depending on the point label value.

[0045] Step 13: Use the Euclidean distance method to calculate the label distance of all vertices between the two graphs, and the obtained label dis...

Embodiment 2

[0063] The method of the second embodiment is substantially the same as that of the first embodiment, the difference is that, in the first embodiment, after calculating the number of cost values ​​that satisfy the constraints in each row and each column, these cost values ​​are replaced by positive infinity at the same time,

[0064] Another graph edit distance method for letter recognition includes the following steps:

[0065] Step 1. Construction of the cost matrix framework of SFBP algorithm.

[0066] Step 11. Use the relevant algorithm to extract the features of any two letters as graph data. Assume that the number of vertices of the two graph data obtained is a and b respectively, each vertex has a digital label, and the edge structure between vertices is undirected and unlabeled. For the convenience of description, according to the number of vertices in the graph, the graphs are called a graph and b graph respectively.

[0067] Step 12. Set the insert operation cost v...

Images