Multilevel Thresholding Technique for Contrast Enhancement in Thermal Images to Facilitate Accurate Image Segmentation

The most challenging task in InfraRed Thermography based condition monitoring of electrical equipment lies in the accurate detection of the anomaly. It necessitates an efficient preprocessing technique for increasing the contrast between the background and the anomaly regions in thermographs. In this paper, an efficient multi-level thresholding based enhancement technique is proposed for highlighting the anomaly region from the background. In contrast to the two levels in binary enhancement, n levels are determined based on the histogram bins. Enhanced intensity is calculated based on the range of the level rather than selecting randomly. Strength of the proposed technique is evident in its ability to detect the true edges corresponding to the anomaly. Also the anomaly region is accurately detected to its true size which in turn resulted in the precise estimation of the real time temperature. The error in temperature is approximately 0.5 at an emissivity of 0.7. The proposed technique can be adapted irrespective of thermal profiling of thermographs. However the intensity range must be chosen appropriately. Relationship between emissivity and error in temperature is also provided at the room temperature