A blur robust color image dectection method based on Maximally Stable Extremal Regions

Maximally Stable Extremal Region (MSER) has been proven to be an excellent feature extraction method in computer vision. However, the initial MSER detector performs not so well in blur situation. Meanwhile, the traditional method is merely applied to intensity space, so that some important color-related information is inevitably overlooked. Hence, this paper utilized the concept of Hysteresis Thresholding to obtain continuous edges information in RGB color channels and extended the original intensity definition and stability criterion of the MSER method in RGB color space with the acquired curves. Result part indicates that in blur and colourful situations, our method performs better compared with the original MSER method and a traditional practice detecting color object based on MSER.