Short user guide

The theoretical fundaments for implementation of this software are explained in Character Recognition with Metasets, Document Recognition and Understanding, INTECH, 2011

The upper part of the application (the menu bar) displays hints explaining the current step. The menu Operations contains the item New which restores the initial state of the program discarding all the changes.

Stage 1: Setting resolution of the matrix

The bottom-left corner of the application window includes the matrix for rendering characters and displaying quality grades. Two sliders placed on the left and at the bottom adjust the dimension of the matrix. The cells of the matrix present nodes of the binary tree assigned to them automatically. The default assignment always uses nodes from one level of the tree. The level is selected based on the number of cells in the matrix, so that all the assigned nodes are pairwise different.

Remark: When the number of cells in the matrix is not a power of 2, then the set of nodes assigned automatically does not constitute a maximal antichain. In such case it is necessary to assign nodes manually. Otherwise the results will be invalid.

Stage 1.a: Assigning nodes to cells of the matrix manually

The bottom-right part of the application presents initial levels of the binary tree. The number of visible levels is adjusted with the slider at the bottom, below the tree view. The sliders on the right and above the tree view allow for scaling the view.

1. Press the left mouse button on a node in the tree, which is to be assigned to a cell of the matrix.
2. Holding the button pressed drag the mouse pointer to the destination cell on the matrix and release the button.
3. The node appears in the destination cell, where the mouse button was released. It replaces the one which was there before.
   1. The selected node is removed from the tree together with all of its descendants.
   2. If the matrix contained nodes comparable to the new one, they are removed from the matrix. Their corresponding cells become empty. Other nodes must be assigned to them.
   The reason for (a) and (b) is the following rule.

   The set of nodes corresponding to cells of the matrix must form an antichain. Also, it must be a maximal antichain, although the program does not force it.

When the assignment of nodes to the cells is complete click the button Define in the bottom-right corner. If there exist empty cells - without any node assigned - then the program displays the error message requiring to fill all the cells. If the assignment is correct the program enters the second stage. The defined assignment of nodes of the tree to cells of the matrix remains constant throughout the pattern matching session. Changing it removes all the pattern elements defined.

The text on the Define button is changed to Configure. Pressing it again all the defined samples are lost and the program enters the stage 1 again.

Stage 2: Defining the compound character pattern

The upper part of the application window contains thumbnails representing the character samples constituting the compound pattern. The current thumbnail is marked with a border. New samples are created with the New button. The Remove button removes the current sample. The buttons Next, Previous navigate through samples. The active sample may be also selected with mouse. The current sample is editable; it is displayed in the lower part of the window.

Initially, one sample is created. It should be edited by selecting pixels as described below. It cannot be removed: there must remain at least one sample in the compound pattern.

The bottom-left corner includes the matrix for defining character samples.

Stage 2.a: Rendering character sample

Use mouse to paint the character on the matrix. The left mouse button selects and deselects cells corresponding to active pixels of the character sample.

Stage 2.b: Evaluating quality degree

The right mouse button selects/deselects nodes in cells which determine the quality degree of the current sample. The cells containing selected nodes are interpreted as valid areas of the character; the remaining deselected cells are the ones which are considered unimportant. The set of selected cells determines the set of nodes of the binary tree which makes up the membership degree of the metaset representing the character in the compound pattern. This set is automatically marked in the tree view on the right. Together with each marked node in the tree also all its descendants are marked (on the visible levels), and - possibly - some parent nodes, according to the forcing relation rules. The numerical value of the membership degree (the quality of the sample) is displayed in the tree view.

The nodes belonging to the quality set may be also chosen directly in the tree view by selecting them with the left mouse button and deselecting with the right button. The set of cells containing nodes comparable to the selected one is marked on the matrix in such case.

The defined character sample, when treated as a metaset, is a potential element of the metaset representing the compound pattern. Its membership degree in the compound pattern is represented by the marked nodes of the tree. The numerical value of the membership is displayed in the tree view as well. The upper part of the application view demonstrates the compound pattern in form of thumbnails corresponding to character samples. In each element in the top panel, above the matrix thumbnail there is a thumbnail of the bottom tree representing the membership (quality) degree. Below the matrix there is a tree thumbnail which will present later the equality (similarity) degree of the testing character sample and the pattern element.

The large button Clear at the bottom resets membership degree visible in the tree view.

Stage 3: Testing samples

When all elements of the compound pattern are defined the program enters the testing stage by pressing the button Test. It changes its functionality then to Define, so that pressing it again goes back to the stage 2 - defining the pattern.

At this stage, the matrix is used for supplying user testing samples. They are painted with mouse similarly as by defining the pattern samples: the left mouse button selects and the right button deselects a cell.

At arbitrary moment the button Check may be pressed to calculate similarity degrees. The main similarity degree of the testing sample to the compound pattern is represented as the membership degree of the metaset representing the testing sample to the metaset representing the compound pattern. It is visible as the set of marked nodes in the tree. Its numerical value is also displayed in the tree view. Additionally, the cells of the matrix containing selected nodes are highlighted. Thus, the remaining (darken) cells demonstrate the parts of the sample which do not match the pattern. They decrease the overall similarity degree. The highlighted cells of the matrix reveal the parts which are considered similar to the pattern and they increase the similarity degree.

In the lower parts of thumbnails in the top panel there appear similarity degrees of the testing sample to each of compound pattern elements, i.e., the equality degrees of the metasets representing them. They are represented as marked nodes of the binary tree and their numerical values are shown too.

Selecting a thumbnail with mouse displays the similarity (equality) degree of the testing sample and the selected one in the main tree view, accompanied by the numerical value. This allows for thorough inspection of the levels of the tree which are invisible in thumbnails.

Please, submit bugs to the author