Computer-Implemented System And Method For Recognizing Patterns In A Digital Image Through Document Image Decomposition

Abstract

A computer-implemented system and method for retrieving a digital image through document image decomposition is provided. A stored digital image is retrieved. Generic visual features are extracted. The features are grouped into a primitive layer including word-graphs that each include words and features. The words are grouped into a layout layer including zone hypotheses that each include one or more of the words. Causal dependencies between the word-graphs and the zone hypotheses are expressed through zone models that include a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier. Each pair of probabilistic models is expressed as an optimal set selection problem including a set of cost functions and constraints. The optimal set selection problem is evaluated through a heuristic search of the cost functions and constraints and a non-overlapping optimal set of the zone hypotheses is provided that characterize the stored digital image.

Description

FIELD

This application relates in general to digital image processing and, in particular, to a system and method for decomposing a digital image.

BACKGROUND

Document processing devices, such as copiers, scanners, and digital facsimile machines, are increasingly able to handle digitally equivalent versions of paper documents, which can contain digitized text and pictorial, graphical, and other data. However, further processing is often needed to convert raw digital images into an electronically usable form, such as needed for pattern recognition, document classification and retrieval, and other tasks. For example, digital images must often be broken down or “decomposed” into constituent parts or “zones.”

Post-digitization image decomposition can be problematic particularly when a large volume of documents are being converted, thereby rendering manual document decomposition impracticable. Conventional page decomposition generally involves bottom-up, top-down, or hybrid methodologies. Bottom-up approaches, such as the Block Adjoining Graph method, detect individual connected components, which are progressively aggregated into higher level structures, such as words, lines, paragraphs, and so forth. Top-down approaches, such as the X-Y Tree method, recursively split a digital image into rectangular areas by alternating horizontal and vertical cuts along white space. These methodologies are typically implemented through ad hoc rules that can be brittle and which often produce varying results, even with little actual change in the data.

Therefore, there is a need for a non-rule based approach to decomposing digital images into constituent parts or zones without a reliance on specific visual aspects, such as connected components, graphical features, and white space.

SUMMARY

A system and method for digital document image processing through heuristic search is provided to decompose digital images into constituent parts or “zones.” Overlapping zone hypotheses are first generated based on generic visual features. Each zone hypothesis, or candidate zone, is evaluated quantitatively by a learned generative zone model, which includes a likelihood model for zones and a prior model for zone relations. An optimal set of non-overlapping zones that covers a given document image are inferred by heuristic searches that can include a binary integer linear programming problem and an A* best-first graph search. Other heuristic searches are possible.

One embodiment provides a computer-implemented system and method for recognizing patterns in a digital image through document image decomposition. A set of exemplary patterns comprised in digital data is maintained. A stored digital image is retrieved from a database of stored documents. Generic visual features are extracted from the stored digital image. The generic visual features are grouped into a primitive layer including a plurality of word-graphs that each include one of words and visualized features. The one or more of the words are grouped into a layout layer including a plurality of zone hypotheses that each include one or more of the words. Causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer are expressed through zone models that each include a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier. Each pair of probabilistic models is expressed as an optimal set selection problem including a set of cost functions and constraints. The optimal set selection problem is evaluated through a heuristic search of the cost functions and constraints and a non-overlapping optimal set of the zone hypotheses is provided that characterize the stored digital image. Each zone hypothesis in the optimal set is matched against the exemplary patterns and the zone hypotheses in the digital image closest matching are identified.

A further embodiment provides a computer-implemented system and method for classifying a digital image through document image decomposition. Identifying characteristics representing documents selected from the group including forms, preprinted documents, textual documents, and hybrid data comprising one or more of pictorial, graphical, and textual data is stored. A stored digital image is retrieved from a database of stored documents. Generic visual features are extracted from the stored digital image. The generic visual features are grouped into a primitive layer including a plurality of word-graphs that each include one of words and visualized features. The one or more of the words are grouped into a layout layer including a plurality of zone hypotheses that each include one or more of the words. Causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer are expressed through zone models that each include a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier. Each pair of probabilistic models is expressed as an optimal set selection problem including a set of cost functions and constraints. The optimal set selection problem is evaluated through a heuristic search of the cost functions and constraints and a non-overlapping optimal set of the zone hypotheses is provided that characterize the stored digital image. Each zone hypothesis in the optimal set is compared to the document identifying characteristics and those documents closest matching are identified.

A still further embodiment provides a computer-implemented system and method for retrieving a digital image through document image decomposition. A stored digital image is retrieved from a database of stored documents. Generic visual features are extracted from the stored digital image. The generic visual features are grouped into a primitive layer including a plurality of word-graphs that each include one of words and visualized features. The one or more of the words are grouped into a layout layer including a plurality of zone hypotheses that each include one or more of the words. Causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer are expressed through zone models that each include a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier. Each pair of probabilistic models is expressed as an optimal set selection problem including a set of cost functions and constraints. The optimal set selection problem is evaluated through a heuristic search of the cost functions and constraints and a non-overlapping optimal set of the zone hypotheses is provided that characterize the stored digital image.

Still other embodiments of the present invention will become readily apparent to those skilled in the art from the following detailed description, wherein are described embodiments by way of illustrating the best mode contemplated for carrying out the invention. As will be realized, the invention is capable of other and different embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and the scope of the present invention. Accordingly, the drawings and detailed description are to be regarded as illustrative in nature and not as restrictive.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram showing a system for decomposing a digital image, in accordance with one embodiment.

FIG. 2 is a process flow diagram showing a method for decomposing a digital image, in accordance with one embodiment.

FIG. 3 is a diagram showing, by way of example, a word-graph.

FIG. 4 is a process flow diagram showing a routine for generating a word-graph for use in the method of FIG. 2.

FIG. 5 is a diagram showing, by way of example, a zone map.

FIG. 6 is a process flow diagram showing a routine for generating a zone map for use in the method of FIG. 2.

FIG. 7 is a process flow diagram showing a routine for determining feature weights for use in the method of FIG. 2.

FIG. 8 is a process flow diagram showing a routine for performing zone inference for use in the method of FIG. 2.

DETAILED DESCRIPTION

System

Digital image decomposition, which is frequently performed within the larger context of image processing, includes identifying the constituent parts or zones of each document page to aid in subsequent processing. FIG. 1 is a block diagram showing a system 10 for decomposing a digital image 14, in accordance with one embodiment. As used herein, the terms “digital image,” “document image,” “digitized document,” and “digitized facsimile” are used interchangeably and refer to a digital image representation of a printed document, which can include digitized text and pictorial, graphical, and other data, either alone or in combination.

In one form of image analysis, digital image decomposition is performed by an image decomposer 11, which operates on a computer system 13, or other image processing device capable of receiving and operating on an electronically-stored digital image, such as copiers, scanners, and digital facsimile machines. Other devices are possible. Additionally, the digital image decomposition can also be performed as a stand alone function apart from image analysis or other tasks.

The image decomposer 11 accepts digital images 14 as inputs and identifies constituent parts or zones 12 within each individual document page, as further described below beginning with reference to FIG. 2 et seq. Digital images 14 can include, for instance, form and preprinted documents 16, such as income tax return forms; textual documents 17, such as publications, news and magazine articles, and the like; and hybrid data 18, such as digital representations of postage stamps, which include pictorial, graphical, and textual data. Other types of digital images 14 are possible.

The image decomposer 11 analyses each digital image 14 and identifies individual zones 12 within each page. Following decomposition, post-analysis operations 15 can be performed, which can include, by way of example, pattern recognition 19, document classification 20, and document retrieval 21. Other operations 15 are possible.

The computer system 13 includes components conventionally found in computing devices, such as a central processing unit, memory, network interface, persistent storage, and network interconnectivity. The image decomposer 11 can also be directly implemented in hardware as a microprogrammed device or dedicated computation module. Other systems and components are possible.

Method

Digital image decomposition identifies zones within each document page. A “zone” is a perceptually compact and consistent block of text, such as a paragraph, or graphical image patch. FIG. 2 is a process flow diagram showing a method 30 for decomposing a digital image 14, in accordance with one embodiment. The method is performed as a series of process steps by general purpose programmable computing device, such as a personal computer, or other image processing device.

Digital image decomposition is performed in representational, modeling, and inferential stages. Initially, each digital image is received as an input in the form of electronically-stored data (step 31), from which logical digital image representations are built (step 32). Each digital image 14 is internally represented as a two-layer hierarchical model based on generic visual features. The hierarchical model includes a “primitive” layer that forms a word-graph (step 33), as further described below with reference to FIGS. 3 and 4, and a “layout” layer that provides overlapping candidate zones or zone “hypotheses” (step 34), as further described below with reference to FIGS. 5 and 6. Causal dependencies occurring between the primitive and layout layers are expressed through zone models (step 35) that are independently scored based on learning. The zone models include a likelihood model for zones (step 36) and a prior model for zone relations (step 37). Zones with poor scores are discarded and zones 12 within the digital image 14 are inferred from the remaining healthy zone hypotheses (step 38), as further described below with reference to FIG. 8. Finally, the identified zones 12 are provided as outputs (step 39). Other processing stages and operations are possible.

Image Representation

Each digital image 14 is represented by a two-layer hierarchical model, which includes a primitive layer and a layout layer.

Word-Graph Generation—the “Primitive” Layer

The primitive layer represents individual “words” as atomic primitives that transform pixels into compact atomic representations. Thus, the “words” need not be lexically-correct words; word fragments are acceptable. The words are connected into a word-graph. FIG. 3 is a diagram showing, by way of example, a word-graph. Other forms of word-graphs are possible.

In terms of notation, the word-graph G_w is denoted:

G_w=V,E  (1)

where N_w is the number of words in the digital image 14; V={v_i;i=1, . . . , N_w} is a set of vertices v in the word-graph G_w that correspond to each word w; and E={(e=(i, j)w_ij): v_iv_jεV, w_ijεR} is a set of edges e that associates a weight w_ij that represents the bounding force between a pair of words w_i and w_j.

FIG. 4 is a process flow diagram showing a routine for generating a word-graph for use in the method 30 of FIG. 2. First, individual “words” within the digital image 14 are found (step 41). The words can be detected through feature extraction through Harr-like filters, also known as Viola-Jones rectangular filters, such as described in P. A. Viola and M. J. Jones, “Rapid Object Detection Using a Boosted Cascade of Simple Features,” Proc. of Comp. Vision and Pattern Recog., pp. 511-518 (2001), the disclosure of which is incorporated by reference. The Han-like filters include horizontal step edge, vertical step edge, horizontal bar, vertical bar, and diagonal block filters, which can be varied in size and scale. Other forms of word detection are possible.

Next, neighboring words are found (step 42). Neighboring words can be identified using a Voronoi tessellation neighbor finding algorithm, such as described in K. Kise et al., “Segmentation of Page Images Using the Area Voronoi Diagram,” Comp. Vision and Image Undemanding, V. 7, No. 3, pp. 370-382 (June 1998), the disclosure of which is incorporated by reference. Other forms of neighbor finding are possible.

The edge weights, including features associated with each word, are then determined (step 44), which are evaluated as posterior probabilities by a learned binary edge classifier (step 43). The features are generic and visual, and include:

• • (1) “Word” compactness in a zone, which is denoted:

$\begin{array}{cc}{f}_w^{\left(0\right)}\left(g\right)=\frac{\sum _{i=1}^kA\left(g_i\right)}{A\left(z\right)}& \left(2\right)\end{array}$

• •  such that 0<f_w⁽⁰⁾(g)≦1, where g_i is the i^th connected component of the word-graph within a zone hypothesis. Generally, k=1 within a zone, which indicates that words are highly connected to one another. A(•) is the area of a connected component bounding box.
  • (2) “Word” height(size) consistency in a zone, which is denoted:

$\begin{array}{cc}{f}_w^{\left(1\right)}\left(g\right)=\frac{n_w^d\left(g\right)}{n_w\left(g\right)}& \left(3\right)\end{array}$

• •  such that 0<f_w⁽⁰⁾(g)≦1, where n_w(g) is the number of words in a zone; and n_w^d(g) is the number of words with dominant height in the zone. This feature provides the ratio of dominant sized words in a zone, which indicates the zone's font size consistency.
  • (3) Zone bounding box top border edge-cutting span, which is denoted:

$\begin{array}{cc}{f}_w^{\left(2\right)}\left(g\right)=\frac{l_e^{\left(t\right)}}{l_z}& \left(4\right)\end{array}$

• •  such that 0≦f_w⁽²⁾(g)≦1, where l_z is the width of a zone, and l_e⁽ⁱ⁾ is the length of the part of a zone bounding box top border that cuts the word-graph's edges.
  • (4) Zonc bounding box bottom border edge-cutting span, which is denoted:

$\begin{array}{cc}{f}_w^{\left(3\right)}\left(g\right)=\frac{l_e^{\left(b\right)}}{l_z}& \left(5\right)\end{array}$

• •  such that 0≦f_w⁽³⁾(g)≦1, where l_z is the width of a zone, and l_e⁽ⁱ⁾ is the length of the part of a zone bounding box bottom border that cuts the word-graph's edges.
  • (5) Zone bounding box vertical border average edge-cutting weight, which is denoted:

$\begin{array}{cc}{f}_w^{\left(4\right)}\left(g\right)=\frac{\sum _{i=1}^{n_e^{\left(v\right)}}w_e^{\left(i\right)}}{n_{tl}}& \left(6\right)\end{array}$

• • where n_e^(v) is the number of edges cut by the two vertical borders is of a zone bounding box; w_e⁽ⁱ⁾ is the i^th edge weight; and n_tl is the number of text lines in the zone. This feature indicates the connection force of a proposed zone with its surroundings. The likelihood of a zone bounding box being a zone decreases as the size of the edge-cutting weight cut increases.
  • (6) Text line alignment in a zone, which is denoted:

f _w ⁽⁵⁾(g)=min(var(x_l),var(x_c),var(x_r))  (7)

• •  This feature provides the minimum variance of the left, center, and right coordinates of a line of text in a zone. The alignment improves as the variance decreases.

These features are heuristic in nature, but are specified independent of language and layout style. The features are not necessarily independent and are utilized to evaluate the “goodness” of proposed zones.

The binary edge classifier is trained (step 43) using word-graphs of training images. During data preparation, zone bounding boxes on the training word-graphs are manually labeled. Edges within zone bounding boxes are labeled positive edge samples, whereas those edges cut by zone bounding box borders are labeled negative samples. Next, a 22-dimensional feature vector is extracted, which includes a feature that accounts for height differences between pairs of words and the 21 Harr-like filter responses detected from an image patch. The image patch is cut by centering the image at the mid-point of an edge over an area that is four times larger than the union of the zone bounding boxes. Finally, the binary edge classifier is trained using the extracted feature vectors, such as described in C. C. Chang and C. J. Lin, “LIBSVM: a Library for Support Vector Machines,” http://www.csie.ntu.edu.tw/˜cjlin/libsvm (2001), the disclosure of which is incorporated by reference. Other types of learned binary edge classifiers are possible.

Zone Hypothesis Generation—the “Layout” Layer

The layout layer groups words into zones, which form a zone map. Each zone is a rectangle, but any polygon representation could be used. FIG. 5 is a diagram showing, by way of example, zone map. Other forms of zone maps are possible.

In terms of notation, the zone map Z is denoted:

Z=(N_z,{z_j:j=1, . . . , N_z})  (8)

where N_z is the number of zones. Each zone z is denoted:

z _j=({c_i^(j):i=1, . . . , n_cj},{v_k^(j):k=1, . . . , n_wj})  (9)

which is a polygon representation, such that c_i^(j) is a corner of a zone bounding polygon; n_cj is the number of vertices or corners of zone-j's bounding polygon; and n_wj is the number of words in zone-j. Although described in terms of a rectangle, the teachings provided herein apply equally to other forms of polygons without loss of generality. Thus, each rectangular zone z is denoted:

z _j=(c_ul,c_lr,{v_k^(j):k=1, . . . , n_wj})  (10)

where c_ul and c_lr are respectively upper-left and lower-right corners of a zone bounding box.

FIG. 6 is a process flow diagram showing a routine 50 for generating a zone map for use in the method 30 of FIG. 2. First, corners of each zone bounding box within the digital image 14 are found (step 52), which can be determined through a learned corner classifier (step 51). Next, all possible zone hypotheses, that is, candidate zones, are proposed by pairing all detected upper-left corners with all lower-right corners (step 53). In a further embodiment, heuristics can be used to propose candidate zones more effectively by ruling out improbable configurations, such as discarding candidate zones that cross line separators.

A pair of corner classifiers to detect upper-left and lower-right corners are trained (step 52) using word-graphs of training images. During data preparation, upper-left and lower-right corners from labeled zones in the training word-graphs are used as positive samples. Negative samples are collected by randomly selecting the corners of word bounding boxes that are not the corners of actual zones. In addition, the corner set of training data by supplemented with bounding box corners of word-graph connected components to help ensure all possible corners are specified. Next, a 21-dimension generic visual feature vector is extracted from an image patch of a predetermined size, which is cut from the image centered at an upper-left or lower-right corner. Finally, a pair of corner classifiers are trained using the extracted feature vectors, such as described in Id. Other types of learned corner classifiers are possible.

Zone Models

From a generative model point of view, the zone hypotheses are causally dependent on the word-graph, which can be expressed as a joint probability:

p(G_w,Z)=p(G_w|Z)p(Z)  (11)

where p(G_w|Z) is a likelihood model for zones, and p(Z) is a prior model for zone relations. Costs and constraints can be assigned to the pair of learned generative zone models, from which non-overlapping zones can be inferred.

Likelihood Model for Zones

The likelihood model for zones for governs how words are organized in zones in terms of the features. The likelihood model for zones p(G_w|Z) can be expressed:

$\begin{array}{cc}p\left(G_wZ\right)=p\left(g_{\stackrel{\_}{w}}\right)\prod _{i=1}^{N_i}p\left(g_iz_i\right)& \left(12\right)\end{array}$

where g_w are sub-graphs of words not covered by any zone, such that p(g_w)=exp(−|g_w|); |•| denotes the cardinality function; g_i is a sub-word-graph subsumed in zone-I; and p(g_i|z_i) is a generative model for zones.

The likelihood model for zones can be constructed as a probabilistic model on word-sub-graphs, such that the expected value of each feature is the same as a corresponding average value extracted from training data. Given n labeled zones, the expected value E can be expressed:

$\begin{array}{cc}\begin{array}{c}{E}_j\left[f_w^{\left(j\right)}\left(gz\right)\right]=\sum _{i=1}^np\left(g_iz_i\right)f_w^{\left(j\right)}\left(g_iz\right)\\ =\frac{1}{n}\sum _{i=1}^nf_w^{\left(j\right)}\left(g_iz_i\right)={\mu }_j\end{array}& \left(13\right)\end{array}$

such that j=0, . . . , 5, where j indexes the zone features. The observed feature statistics serve as constraints and based on the maximum entropy principle, the likelihood model for zones can be derived as:

$\begin{array}{cc}p\left(gz\right)=c\exp \left\{-\sum _{j=0}^5{\lambda }_jf_w^{\left(j\right)}\left(gz\right)\right\}& \left(14\right)\end{array}$

where the λ's are Lagrange multipliers or, more particularly, feature weights to be estimated; and c is a normalizing constant. As features f_w⁽²⁾, f_w⁽³⁾, and f_w⁽⁴⁾ are context sensitive, the likelihood model for zones also encodes a certain amount of contextual information.

The feature weights are solved iteratively using Generalized Iterative Scaling (GIS), such as described in J. N. Darroch and D. Rateli, “Generalized Iterative Scaling for Log-Linear Models,” Ann. Of Math. Stats., V. 43 (1972), the disclosure of which is incorporated by reference. FIG. 7 is a process flow diagram showing a routine for determining feature weights 60 for use in the method 30 of FIG. 2. Initially, given n labeled zones, each feature of each zone is determined (step 61), such that f_w^(j)(g_i|z_i), where (j=0, . . . , 5, i=1, . . . , n). The average of each feature extracted from the training data is found (step 62):

$\begin{array}{cc}{\mu }_j=\frac{1}{n}\sum _{i=1}^nf_w^{\left(j\right)}\left(g_iz_i\right)& \left(15\right)\end{array}$

where j=0, . . . , 5. To initialize the GIS, an initial feature weight λ for each feature is set (step 63):

λ_j⁽⁰⁾=1  (16)

where j=0, . . . , 5. During each iteration t of the GIS, an expected value of each feature is evaluated with the current feature weight (step 64), expressed:

$\begin{array}{cc}{E}_j^{\left(t\right)}\left[f_w^{\left(j\right)}\left(gz\right)\right]=\sum _{i=1}^np^{\left(t\right)}\left(g_iz_i\right)f_w^{\left(j\right)}\left(g_iz_i\right)& \left(17\right)\end{array}$

where j=0, . . . , 5. The feature weight for each feature is then updated (step 65):

$\begin{array}{cc}{\lambda }_j^{\left(t+1\right)}={\lambda }_j^{\left(t\right)}+\frac{1}{C}\log \frac{{\mu }_j}{E_j^{\left(t\right)}}& \left(18\right)\end{array}$

where j=0, . . . , 5; C is a correction constant chosen large enough to cover an additional dummy feature, such as described in Id. In one embodiment, C=8. Iteration of the GIS continues until convergence (step 66).

Prior Model for Zone Relations

The prior model for zone relations governs each zone's shape and the spatial distribution of zones within a document page, including similarity, proximity, symmetry, and other attributes. The prior model for zone relations is characterized by a Gestalt ensemble for various Gestalt patterns, such as described in C. Guo et al., “Modeling Visual Patterns by Integrating Descriptive and Generative Methods,” Int'l J. of Comp. Vision, V. 53, No. 1, pp. 5-29 (June 2003), the disclosure of which is incorporated by reference. This model makes zone evaluation context sensitive. To minimize learning expense, the specificity of the document set is used opportunistically by requiring that each zone be a rectangle with no overlap between any two zones, such that:

p({z₁, . . . , z_Nz})=Π_i≠jδ(z_i∩z_j)  (19)

where δ(•) is a Dirac delta function. The prior model for zone relations p(Z) can be expressed:

$\begin{array}{cc}p\left(Z\right)=p\left(N_z\right)\prod _{i\ne j}^{}\delta \left(z_i\bigcap z_j\right)& \left(20\right)\end{array}$

where p(N_z) is prior knowledge on zone cardinality, which is assumed a uniform distribution.

Thus, the joint probability of a word-graph G_w and zone partition Z is:

$\begin{array}{cc}\begin{array}{c}p\left(G_w,Z\right)=p\left(G_wZ\right)p\left(Z\right)\\ =p\left(g_{\stackrel{\_}{w}}\right)\left\{\prod _{i=1}^{N_i}p\left(g_iz_i\right)\right\}·p\left(N_z\right)\prod _{i\ne j}^{}\delta \left(z_i\bigcap z_j\right)\end{array}& \left(21\right)\end{array}$

Zone Inference

The goal of document image decomposition is to segment a given document image into an optimal set of zones, such that the words in each zone are well organized, pairs of zones do not overlap, the full set of zones covers all of the words in the document, and the number of segmented zones conforms with prior knowledge. An optimal set of zones is inferred by converting learned data statistics into costs and constraints. FIG. 8 is a process flow diagram showing a routine for performing zone inference 70 for use in the method 30 of FIG. 2. The foundational aspects of inferring zones are described generally and specific forms of heuristic searches are described in detail by way of examples that include a binary integer linear programming problem and an A* best-first graph search. Other heuristic searches are possible.

Initially, a set of cost functions and constraints are defined to covert the statistical inference into an optimal set selection problem (step 71). Both the likelihood model for zones and the prior model for zone relations are learned probabilistic models. A probability 0<P(•)<1 can be converted into a cost as:

c(•)=ρ(−log P(•))  (22)

where ρ(x) is a robust function for truncating extreme values. When P(•)=0 or P(•)=1, a binary constraint is generated. Accordingly, the learned probabilistic models generate the following cost functions and constraints: (1) an individual cost for each zone; (2) a binary constraint that selected zones cover all words in a page; and (3) a binary constraint requiring no overlap between any pair of zones.

The remaining operations, formulating heuristic search parameters (step 72) and evaluating the heuristic search (step 73) are heuristic-search dependent.

Binary Integer Linear Programming Problem

Once all possible zone hypotheses have been proposed, page decomposition can be expressed as an optimal set selection problem, which can be modeled as a binary integer linear programming (IPL) problem that can be passed to a standard IPL solver, such as CPLEX. Other IPL solvers are possible.

The binary IPL problem can be formulated by first introducing n binary variables x_i. Accordingly, given n candidate zones, x is a binary state vector, such that x=(x₁, . . . , x_n), such that x_i=1 means zone-i is selected and x_i=0, otherwise. Any specific choice of 0's or 1's for the components of x corresponds to selecting a particular subset of candidate zones. Thus, every choice of x represents a two-way partition of the set of zones between those zones that have been selected, and those zones that have not. The IPL problem parameters are formulated (step 72) in the context of document decomposition and evaluated (step 73):

• • (1) Total zone cost of the selection corresponding to x, defined as:

C _z(x)=c_z^Tx  (23)

• •  where c_z=(c_z,1, . . . , c_z,n)^T is a vector of individual zone costs determined only once according to equations (14) and (22) immediately after the candidate zones are proposed.
  • (2) Prior cost of total number of zones expected, defined as:

C _N(x)=c_N^TΔ(1^Tx)  (24)

• •  where c_N=(c_N,1, . . . , c_N,m)^T is a vector, each of whose entries is the cardinality cost of zone-i, which is obtained from training data p(N_z) and equation (22). Generally, 1≦m≦n and 1<N_z≦m. Δ(1^Tx)=(δ(1^Tx−1), . . . , δ(1^Tx−N_z))^T, where δ(•) is a Dirac delta function. C_N(x) returns a corresponding cost of the selected zone cardinality. However, the cost is not linear due to the Dirac delta function. Another set of binary variables y=(y₁, . . . , y_m) and two additional constraints are introduced to make the cost linear:

C _N(y)=c_N^Ty

such that, 1^Ty=1

1^Tx=I^Ty  (25)

• •  where I=(1, . . . , m)^T is an index vector.
  • (3) Non-overlap constraint about each pair of zones: an n×n Boolean matrix M indicating pairwise zone overlaps is constructed immediately after the candidate zones are proposed. For each entry M(i, j), if TRUE, the i^th zone overlaps with the j^th zone; FALSE indicates no overlap. Only the upper left corner of matrix M is evaluated due to the symmetry of the matrix. Moreover, whenever M(i, j) is TRUE, the constraint x_i+x_j≦1 is enforced. The linear inequality constraints can be compactly expressed:

A_ox≦1  (26)

• •  where matrix A_o is generated from matrix M, such that for each M(i, j) that is TRUE, a row of zeroes is added, except that a ‘1’ is placed in the i^th and j^th locations. 1 is the vector of all ones of appropriate size.
  • (4) Word covering constraint about a set of selected zones: Given N_w-word page, the selected set of zones should follow the constraint:

n_wz^Tx=N_w  (27)

• •  where n_wz=(n_w,1, . . . , n_w,n)^T and N_w,i are the number of words in the i^th candidate zone, which is only determined once immediately after the candidate zones are proposed.

Based on the IPL problem formulation, the goal can be expressed as:

minimize, C_z(x)+C_N(y)

such that, 1^Ty=1

1^Tx=I^Ty

A_ox≦1

n_wz^Tx=N_w

x_i≈{0,1}, y_iε{0,1}

I=(1, . . . , m)^T  (28)

Finally, applying equations (23) and (25), equation (23) can be rewritten as an integer linear program:

minimize, c_N^Tx+c_N^Ty

such that, 1^Ty=1

1^Tx=1^Ty

A_ox≦1

n_wz^Tx=N_w

x_iε{0,1}, y_iε{0,1}

I=(1, . . . , m)^T  (29)

A* Best-First Graph Search

The A* search is a best-first graph search algorithm, which finds a path from an initial node to a goal node. The search maintains a set of partial solutions that are paths through the graph, starting at an initial node, and which are stored in a priority queue. The priority assigned to a path passing through a node x is determined by the function:

f(x)=g(x)+h(x)

where g(x) is a cost function that measures the cost incurred from the initial node to the current node x; and h(x) is a heuristic function that estimates the cost from the current node x to the goal node. To ensure the search finds an optimal solution, h(x) must be admissible.

After zone hypotheses are proposed, page decomposition can be formulated as a weighted polygon partitioning problem in computational geometry: given a polygon (a document page) and a set of candidate sub-polygons (zones), each zone having an associated weight (cost), the goal becomes partitioning the polygon into a subset of disjoint sub-polygons from the candidate set, so as to cover every word in a document image with minimum cost. This page decomposition problem can be solved by an A* search, which exploits heuristics from data to improve search performance.

The A* search operates on a set of parameters, which are formulated (step 72) in the context of document decomposition and evaluated (step 73):

• • (1) State Variable x: Given n candidate zones, x is a binary state vector, such that x=(x₁, . . . , x_n), such that x_i=1 means zone-i is selected and x_i=0, otherwise. Any specific choice of 0's or 1's for the components of x corresponds to selecting a particular subset of candidate zones.
  • (2) Goal State: the goal state is that every word in a given document is covered by only one zone.
  • (3) Cost Function: The cost of each path to x is defined as:

g(x)=c_z^Tx  (29)

• •  where c_z=(c_z1, . . . , c_zn)^T is a vector of individual zone costs determined only once according to equations (14) and (22) immediately after the candidate zones are proposed.
  • (4) Heuristic Function: to insure that the A* search is admissible or optimal, the heuristic function h(x) must not overestimate the actual cost of reaching the goal state. To ensure admissibility, h(x) for a path from x to the goal state, the minimum number of non-overlapping polygons to partition the rest of the document page that has not been covered by the selected zones, n_z(x), is estimated, assuming that both the document and the zones are represented by rectangles, using the expression:

h(x)=n_z(x)*c_min  (30)

• •  where c_min is the minimum zone cost learned from the training data. The estimate of n_z(x) involves partitioning the complementary polygon of a state x, which is created by removing the selected zones (rectangles) from the document page (a rectangle) to form a minimum number of non-overlapping rectangles, restricted to only orthogonal polygons whose edges are either horizontal or vertical, such as described in W. Lipski et al., “On Two Dimensional Data Organization II,” Fundamental Informaticae, V. 2, pp. 227-243 (1977), the disclosure of which is incorporated by reference.

While the invention has been particularly shown and described as referenced to the embodiments thereof, those skilled in the art will understand that the foregoing and other changes in form and detail may be made therein without departing from the spirit and scope.

Claims

1. A computer-implemented method for recognizing patterns in a digital image through document image decomposition, comprising: maintaining a set of exemplary patterns comprised in digital data;extracting generic visual features from a stored digital image;grouping the generic visual features into a primitive layer comprising a plurality of word-graphs that each comprise one of words and visualized features;grouping the one or more of the words into a layout layer comprising a plurality of zone hypotheses that each comprise one or more of the words;expressing causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer through zone models that each comprise a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier;expressing each pair of probabilistic models as an optimal set selection problem comprising a set of cost functions and constraints;evaluating the optimal set selection problem through a heuristic search of the cost functions and constraints and providing a non-overlapping optimal set of the zone hypotheses that characterize the stored digital image; andmatching each zone hypothesis in the optimal set against the exemplary patterns and identifying the zone hypotheses in the digital image closest matching.

2. A method according to claim 1, further comprising: detecting each of the words in the layout layer through Harr-like filters selected from the group comprising horizontal step edge, vertical step edge, horizontal bar, vertical bar, and diagonal block filters, which can each be varied in size and scale.

3. A method according to claim 2, further comprising: identifying one or more words that neighbor each detected word through a Voronoi tessellation neighbor finding algorithm.

4. A method according to claim 1, further comprising: determining a likelihood model for each zone hypothesis in each zone model comprising a probabilistic model on word-sub-graphs, wherein an expected value of each generic visual feature is the same as a corresponding average value of a generic visual feature extracted from training data.

5. A method according to claim 1, further comprising: determining a prior model for zone relations in each zone model relating to a shape and spatial distribution of the zone hypotheses within the stored digital image characterized by a Gestalt ensemble.

6. A method according to claim 1, further comprising: performing the heuristic search as a binary integer linear programming problem comprising minimizing the total cost function of each zone hypothesis and the linear prior cost of the total number of zone hypotheses expected.

7. A method according to claim 1, further comprising: performing the heuristic search as an A* search comprising a path from an initial node to a goal node through the set of zone hypotheses, wherein the initial node comprises a current zone hypothesis and the goal node comprises a partition of the zone hypotheses set comprising a subset of disjoint sub-zone hypotheses.

8. A computer implemented method for classifying a digital image through document image decomposition, comprising: storing identifying characteristics representing documents selected from the group comprising forms, preprinted documents, textual documents, and hybrid data comprising one or more of pictorial, graphical, and textual data;extracting generic visual features from a stored digital image;grouping the generic visual features into a primitive layer comprising a plurality of word-graphs that each comprise one of words and visualized features;grouping the one or more of the words into a layout layer comprising a plurality of zone hypotheses that each comprise one or more of the words;expressing causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer through zone models that each comprise a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier;expressing each pair of probabilistic models as an optimal set selection problem comprising a set of cost functions and constraints;evaluating the optimal set selection problem through a heuristic search of the cost functions and constraints and providing a non-overlapping optimal set of the zone hypotheses that characterize the stored digital image; andcomparing each zone hypothesis in the optimal set to the document identifying characteristics and identifying those documents closest matching.

9. A method according to claim 8, further comprising: detecting each of the words in the layout layer through Harr-like filters selected from the group comprising horizontal step edge, vertical step edge, horizontal bar, vertical bar, and diagonal block filters, which can each be varied in size and scale.

10. A method according to claim 9, further comprising: identifying one or more words that neighbor each detected word through a Voronoi tessellation neighbor finding algorithm.

11. A method according to claim 8, further comprising: determining a likelihood model for each zone hypothesis in each zone model comprising a probabilistic model on word-sub-graphs, wherein an expected value of each generic visual feature is the same as a corresponding average value of a generic visual feature extracted from training data.

12. A method according to claim 8, further comprising: determining a prior model for zone relations in each zone model relating to a shape and spatial distribution of the zone hypotheses within the stored digital image characterized by a Gestalt ensemble.

13. A method according to claim 8, further comprising: performing the heuristic search as a binary integer linear programming problem comprising minimizing the total cost function of each zone hypothesis and the linear prior cost of the total number of zone hypotheses expected.

14. A method according to claim 8, further comprising: performing the heuristic search as an A search comprising a path from an initial node to a goal node through the set of zone hypotheses, wherein the initial node comprises a current zone hypothesis and the goal node comprises a partition of the zone hypotheses set comprising a subset of disjoint sub-zone hypotheses.

15. A computer-implemented method for retrieving a digital image through document image decomposition, comprising: retrieving a stored digital image from a database of stored documents;extracting generic visual features from the stored digital image;grouping the generic visual features into a primitive layer comprising a plurality of word-graphs that each comprise one of words and visualized features;grouping the one or more of the words into a layout layer comprising a plurality of zone hypotheses that each comprise one or more of the words;expressing causal dependencies between the word-graphs in the primitive layer and the zone hypotheses in the layout layer through zone models that each comprise a joint probability defining a pair of probabilistic models generated through a learned binary edge classifier;expressing each pair of probabilistic models as an optimal set selection problem comprising a set of cost functions and constraints;evaluating the optimal set selection problem through a heuristic search of the cost functions and constraints and providing a non-overlapping optimal set of the zone hypotheses that characterize the stored digital image.

16. A method according to claim 15, further comprising: detecting each of the words in the layout layer through Harr-like filters selected from the group comprising horizontal step edge, vertical step edge, horizontal bar, vertical bar, and diagonal block filters, which can each be varied in size and scale.

17. A method according to claim 16, further comprising: identifying one or more words that neighbor each detected word through a Voronoi tessellation neighbor finding algorithm.

18. A method according to claim 15, further comprising: determining a likelihood model for each zone hypothesis in each zone model comprising a probabilistic model on word-sub-graphs, wherein an expected value of each generic visual feature is the same as a corresponding average value of a generic visual feature extracted from training data.

19. A method according to claim 15, further comprising: determining a prior model for zone relations in each zone model relating to a shape and spatial distribution of the zone hypotheses within the stored digital image characterized by a Gestalt ensemble.

20. A method according to claim 15, further comprising: performing the heuristic search as a binary integer linear programming problem comprising minimizing the total cost function of each zone hypothesis and the linear prior cost of the total number of zone hypotheses expected.

21. A method according to claim 15, further comprising: performing the heuristic search as an A search comprising a path from an initial node to a goal node through the set of zone hypotheses, wherein the initial node comprises a current zone hypothesis and the goal node comprises a partition of the zone hypotheses set comprising a subset of disjoint sub-zone hypotheses.