A Two-step Thresholding Algorithm for Image-based Defect Detection in Roll-to-Roll Coating of Cellulose Nanocrystal Films

Roll-to-roll gravure coating is a scalable process to manufacture cellulose nanocrystals (CNC) based films for packaging and printing. In developing effective roll-to-roll (R2R) process control for coating CNC, real-time inline defect detection is needed. In particular, in many image analysis and processing algorithms, automated thresholding is a fundamental step. Otsu's method is one of the most commonly used techniques for automatic thresholding. It is based on a key assumption that the image exhibits a multimodal histogram distribution. CNC coated films have different textures and the associated defects usually occupy a very small portion of the captured image. This results in an image that is difficult for Otsu's method to generate desirable thresholds. In this work, a two-step automatic thresholding algorithm is proposed, where a statistical process control (SPC) based algorithm is employed to determine a preliminary threshold value for removing majority of the background pixels. This will accentuate the pixel-to-pixel variations that will facilitate the use of Otsu's method to determine the final threshold to isolate fine defect regions. Comparative experimental results verified the effectiveness of the proposed method in finding thresholds that segment out defect pixels from different backgrounds and textures associated with roll-to-roll processed CNC films.