Robust image matching for irregular illumination variation based on spatio-temporal analysis of image intensity

In this paper, we proposed a high-speed and robust image matching method for illumination variation. Our method selects a small number of temporally stable pixels. Such pixels realize both high-speed matching and robustness for illumination changing. The previous SCPTM method has an important problem which cannot suppress irregular illumination changes. The reason is that the SCPTM uses some images obtained in fixed regular intervals for estimating stability of intensity in image sequence. In this research, we estimate stability of pixel based on trend analysis of intensity of matching region from every time. The trend analysis of image intensity is to analyze the temporal changes of each pixel in a series of multiple images. The key idea of this research is to evaluate temporal stability of the pixel by investigating the similarity of the trend. Sequential information of intensity trend makes it possible to estimate stability of pixel in irregular illumination variation. Both robustness for illumination variation and high-speed are achieved by selecting robust pixels for irregular variation. It has been proved that our system achieves 99.8% recognition success rate through an experiment with 21600 test images including drastic illumination variation such as west sun in a fine day. Also, we have shown that it can be utilized for detection of various kind of object, such as IC parts, metal parts and small printed labels. Recognition success rate of the proposed method is approximately 5% higher than our previous work (SCPTM method).