Image analysis based on location sampling

Abstract

An image is analyzed by random or pseudorandom sampling. Information about attributes such as skew, typeface, and character size and spacing is obtained by measuring a characteristic at a number of sampled locations and combining the results to obtain a measure of the characteristic for the image. The number of sampled locations is smaller than the number of locations in the image, but is large enough to achieve a degree of statistical significance. The sampled locations can be selected by obtaining a random or pseudorandom number indicating a candidate location and then applying a criterion to determine whether the candidate location is an appropriate starting point for the measurement. For example, the criterion can be that the location must be a pixel within a character or must be a pixel of a given color at an edge. The measurement can indicate a distance from each sampled location in each of a number of directions, such as the distance to an edge. The distances can be combined to obtain data indicating a distribution, such as by averaging. An image can be divided into segments, each of which has a prominent value for a characteristic being measured, by first dividing the image into starting segments, then sampling the starting segments to measure the characteristic, and then grouping the starting segments into larger segments based on the measurements. The starting segments can be rectangles or, for an image with words, can be segments that each include a word or a paragraph.