Generating processing sequences for image-based decision systems
View Patent ↗A multilevel Chain-And-Tree (CAT) model provides a framework to facilitate highly effective image analysis and measurement. Methods are described to automatically or semi-automatically create a processing sequence for image based decision applications using the CAT model framework. The invention supports new image based decision applications with little or no involvement of image processing experts. A new representation for processing sequence is consistent with CAT data representation, which allows easy understanding, updating, and debugging.
1. An object detection learning method comprises the following steps:
(a) inputting a reference CAT model containing the basic structure and component feature values of an ideal subject and its tolerance ranges wherein each node of the reference CAT model specifies its expected CAT component types, expected CAT component features and expected CAT component relations and reference mask;
(b) inputting at least one learning image;
(c) inputting a detection algorithm database containing multiple detection algorithms for incorporation into the detection sequence of a CAT node;
(d) performing candidate algorithm generation using the detection algorithm database based on the type of component, component features and component relations stored in the reference CAT model having an candidate algorithm output;
(e) performing candidate algorithm evaluation using the candidate algorithm output to select the appropriate detection algorithm for the component of interest having a detection algorithm output wherein the candidate algorithm evaluation step further comprises the following steps:
(i) performing candidate algorithm application using the at least one learning image and the algorithm candidate output having an algorithm result output containing detected component mask;
(ii) performing candidate algorithm scoring using the reference CAT model and the algorithm result output containing detected component feature values having an algorithm score output;
(iii) performing detection algorithm selection using the algorithm score output having a detection algorithm output.
2. The method of claim 1 wherein the candidate algorithm scoring step uses the reference mask stored in the reference CAT model and the detected component mask in the algorithm result output to generate a mask score in the algorithm score output.
3. The method of claim 1 wherein the candidate algorithm scoring step uses the component feature values stored in the reference CAT model and the detected component feature values in the algorithm result to generate at least one feature score in the algorithm score output.