Fast object localization using multi-scale image relevance function

An object detection method using a model-based visual attention mechanism is proposed in application to visual inspection problems and medical diagnostic imaging. The proposed method is based on a multi-scale operator called image relevance function - a non-linear multi-scale filter bank that has local maxima at the centers of object locations. Model-based design of this operator takes into account intensity, shape and texture features of the objects to be detected. This approach offers several advantages, including fast and accurate object localization, simple extraction of shape features, adaptive segmentation of object regions.