Page decomposition using local orthogonal transforms and a map optimization
View Patent ↗Document processing and understanding can be important for a variety of applications such as office automation, creation of electronic manuals, online documentation and annotation, etc. The first step towards this process can involve the decomposition of a page into separate units describing the text, graphics, pictures etc. Towards that end, a method of page decomposition comprises segmenting an image into a plurality of blocks according to an amount of information contained in each block, classifying each block according to a corresponding frequency spectrum, and refining a classification of each block.
1. A computer-implemented method of page decomposition comprising the steps of:
segmenting an image into a plurality of blocks according to an amount of information contained in each block, wherein the segmenting comprises;
dividing, iteratively, the plurality of blocks into a plurality of child blocks, wherein each of the plurality of blocks has a set of corresponding child blocks;
determining an entropy for each of the plurality of blocks and the set of corresponding child blocks; and
performing a further iteration of the dividing on at least one of the plurality of blocks upon determining the entropy of the at least one of the plurality of blocks is greater than the entropy of the set of corresponding child blocks;
classifying each block as one of a text, an image and a graphics block according to a corresponding frequency spectrum and a corresponding value of a neighborhood; and
refining a classification of each block by iterating the classifying of each block.
2. The method of claim 1 , further comprising the steps of:
folding each block along at least one edge, wherein each block comprises a plurality of edges, wherein folding further comprises applying a rising cutoff function to each folded block; and
determining a discrete cosine transform of each block prior to determining the entropy of the block.
3. The method of claim 1 , wherein the step of classifying each block further comprises the steps of:
determining, for each block a plurality of probabilities according to the frequency spectrum of each block, wherein each probability corresponds to a content type;
determining for each block a maximum a posteriori probability; and
classifying each block within a Bayesian-framework comprising block values and neighborhood values.
4. The method of claim 3 , wherein the frequency spectrum of each block is classified according to a concentration of frequency content and variance of the frequency spectrum.
5. The method of claim 1 , wherein an initial segmentation of the image is performed according to a predefined block size.
6. The method of claim 5 , wherein the predefined block size is 256 pixels by 256 pixels.
7. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for page decomposition, the method steps comprising:
segmenting an image into a plurality of blocks according to an amount of information contained in each block, wherein the segmenting comprises;
dividing, iteratively, the plurality of blocks into a plurality of child blocks, wherein each of the plurality of blocks has a set of corresponding child blocks;
determining an entropy for each of the plurality of blocks and the set of corresponding child blocks; and
performing a father iteration of the dividing on at least one of the plurality of blocks upon determined the entropy of the at least one of the plurality of blocks is greater than the entropy of the set of corresponding child blocks;
classifying each block as one of a text, an image and a graphics block according to a corresponding frequency spectrum; and
refining a classification of each block.
8. The method of claim 7 , further comprising the steps of:
folding each block along at least one edge, wherein each block comprises a plurality of edges, wherein folding further comprises applying a rising cutoff function to each folded block; and
determining a discrete cosine transform of each block prior to determining the entropy of the block.
9. The method of claim 7 , wherein the step of classifying each block further comprises the steps of:
determining, for each block a plurality of probabilities according to the frequency spectrum of each block, wherein each probability corresponds to a content type;
determining for each block a maximum a posteriori probability; and
classifying each block within a Bayesian-framework comprising block values and neighborhood values.
10. The method of claim 9 , wherein the frequency spectrum of each block is classified according to a concentration of frequency content and variance of the frequency spectrum.
11. The method of claim 7 , wherein an initial segmentation of the image is performed according to a predefined block size.
12. The method of claim 11 , wherein the predefined block size is 256 pixels by 256 pixels.
13. A system adapted to decompose a page comprising:
a segmenting module for excepting an image, and dividing the image into a plurality of blocks according to an entropy of each block, wherein the segmenting module divides, iteratively, the plurality of blocks into a plurality of child blocks, wherein each of the plurality of blocks has a set of corresponding child blocks, determines an entropy for each of the plurality of blocks and the set of corresponding child blocks, and performs a further iteration of the dividing on at least one of the plurality of blocks upon determining the entropy of the at least one of the plurality of blocks is greater than the entropy of the set of corresponding child blocks; and
a classifying module coupled to the segmenting module for excepting the plurality of blocks, wherein each block is classified as one of a text, an image and a graphics block according to a concentration component, a variance component of the frequency spectrum, and a neighborhood value, and wherein the classifying module iterates through the plurality of blocks and determines a percentage of classification changes of the plurality of blocks from a prior iteration to a threshold, and upon determining the percentage to be at or below the threshold outputting a resulting classified image comprising a plurality of classified blocks.
14. The method of claim 4 , wherein the classifying of each block is performed according to the corresponding frequency spectrum and a corresponding value of a neighborhood, and the refining of the classification of each block is performed by iterating the classifying of each block.
15. The method of claim 14 , further comprising comparing, at each iteration, a percentage of classification changes from a prior iteration to a threshold, and upon determining the percentage to be at or below the threshold, outputting a classified image.
16. The method of claim 1 , further comprising comparing, at each iteration, a percentage of classification changes from a prior iteration to a threshold, and upon determining the percentage to be at or below the threshold, outputting a classified image.