A method of inspecting semiconductor die and lead frame assemblies uses rotation invariant/scale invariant processing methods of machine vision data. A training image is acquired and processed to form a training model. A runtime image is acquired and processed using rotation invariant/scale invariant tools to find a runtime instance of the trained model and produce x, y, theta and scale information. The runtime instance is aligned to the train time model, or vice versa, and then compared to the train time model. The features and edges from the runtime image are compared to features and edges of the training model to identify discrepancies as possible defects. The possible defects are further processed with a morphological filter and/or a blob filter to further refine images of the defects. Alternative implementations of the invention measure adhesive wet-out around semiconductor dies and provide measurements of die rotation.
展开▼
