Search

Computer vision software is complex, involving many tens of thousands of lines of code. Coding mistakes are not uncommon. When the vision algorithms are run on controlled data which meet all the algorithm assumptions, the results are often statistically predictable. This renders it possible to statistically validate the computer vision software and its associated theoretical derivations. In this paper, we review the general theory for some relevant kinds of statistical testing and then illustrate this experimental methodology to validate our building parameter estimation software. This software estimates the 3D positions of buildings vertices based on the input data obtained from multi-image photogrammetric resection calculations and 3D geometric information relating some of the points, lines and planes of the buildings to each other.

Noise models are crucial for designing image restoration algorithms, generating synthetic training data, and predicting algorithm performance. There are two related but distinct estimation scenarios. The first is model calibration, where it is assumed that the input ideal bitmap and the output of the degradation process are both known. The second is the general estimation problem, where only the image from the output of the degradation process is given. While researchers have addressed the problem of calibration of models, issues with the general estimation problems have not been addressed in the literature. In this paper, we describe a parameter estimation algorithm for a morphological, binary, page-level image degradation model. The inputs to the estimation algorithm are 1) the degraded image and 2) information regarding the font type (italic, bold, serif, sans serif). We simulate degraded images using our model and search for the optimal parameter by looking for a parameter value for which the local neighborhood pattern distributions in the simulated image and the given degraded image are most similar. The parameter space is searched using a direct search optimization algorithm. We use the p-value of the Kolmogorov-Smirnov test as the measure of similarity between the two neighborhood pattern distributions. We show results of our algorithm on degraded document images.

Image segmentation is an important component of any document image analysis system. While many segmentation algorithms exist in the literature, very few i) allow users to specify the physical style, and ii) incorporate user-specified style information into the algorithm's objective function that is to be minimized. We describe a segmentation algorithm that models a document's physical structure as a hierarchical structure where each node describes a region of the document using a stochastic regular grammar. The exact form of the hierarchy and the stochastic language is specified by the user, while the probabilities associated with the transitions are estimated from groundtruth data. We demonstrate the segmentation algorithm on images of bilingual dictionaries.