Method and apparatus for automatic image segmentation using template matching filters

A method performed in an digital processor for processing a document image to determine image types present therein, the steps comprising:

• receiving, from an image source, a document image having a plurality of pixels therein, each pixel represented by a density signal, and storing at least a portion thereof representing a region of the document image in a data buffer;
• retrieving, from the data buffer, the density signals for the document image;
• and
• determining, using template matching filters, image types present in the region of the document image by applying a first stage template matching filter to the density signals for the document image to produce a first filtered output image, said step of applying a first stage template matching filter to the density signals for the document image comprising
• identifying a window within the document image so as to select a subset of the document image density signals and comparing the density signals in the window to a pattern within the template based filter, the pattern representing a pattern determined to occur in a textual segment of an image wherein a plurality of signals representing a pattern within the template-based filter are stored in each location of a look-up table memory and the step of comparing the density signals in the window to a pattern within the template-based filter comprises
• storing the subset of document image density signals in a register memory,
• logically comparing each density signal in the register memory with a uniquely associated signal of the template based filter stored in the look-up table memory location, and
• outputting a logic signal indicative of the result obtained in the logical comparing step
• applying a second stage template matching filter to the first filtered output image to produce an error image, and XORing the error image and the first filtered output image to mask from the first filtered output image any density signals from segments of the document image not comprised of text to produce a first output image, wherein the first output image contains only textual segments therein.

An apparatus for processing binary image pixels in an image represented by a plurality of rasters of pixels, to preferentially pass regions having a first structure therethrough so as to produce a binary output image primarily comprised of regions exhibiting the first structure, including:

• an image memory for storing the binary image signals;
• a window buffer for storing a plurality of image signals from a plurality of rasters, said image signals representing pixels centered about a target pixel;
• a template filter to generate the binary output image signal as a function of the image signals stored in the window buffer, wherein the binary output signal is equivalent to the image signal for regions of the binary image where the target pixel represents the first structure, and where the binary output signal is zero for regions of the binary image where the target pixel matches another structure, said template filter thereby preferentially passing target pixels having the first structure;
• and
• an output memory for storing the binary output signal for each of a plurality of target pixels, wherein the signals stored in each location of said output memory are generated by said template filter as a function of the image signals within a window whose contents are determined as a function of the corresponding target pixel location.

An apparatus for processing binary image pixels in an image represented by a plurality of rasters of binary image pixels, to identify regions exhibiting a particular structure therein, comprising:

• an image source for producing a document image having a plurality of binary image pixels therein, each pixel represented by a binary density signal;
• memory for storing at least a portion of the binary density signals representing a region of the document image in a data buffer;
• and
• a segmentation circuit employing template-matching filters to identify the presence of the particular structure in the region of the image stored in said memory, the segmentation circuit further comprising a logic filter for removing the particular structure from the region of the image stored in said memory to produce an output image substantially void of the particular structure;
• wherein the template-matching filters minimize classification error only for the particular structure, while exhibiting greater classification errors for another structure.

An apparatus for processing binary image pixels in an image represented by a plurality of rasters of binary image pixels, to identify regions exhibiting a particular structure therein, comprising:

• an image source for producing a document image having a plurality of binary image pixels therein, each pixel represented by a binary density signal;
• memory for storing at least a portion of the binary density signals representing a region of the document image in a data buffer;
• and
• a segmentation circuit employing template-matching filters to identify the presence of the particular structure in the region of the image stored in said memory, the segmentation circuit further comprising a logic filter for removing the particular structure from the region of the image stored in said memory to produce an output image substantially void of the particular structure;
• wherein said segmentation circuit further comprises a second logic filter for preferentially passing the particular structure for the image region stored in said memory to produce an output image substantially comprised of regions containing the particular structure.