Adaptive Visual Inspection Method for Transparent Label Defect Detection of Curved Glass Bottle

Automatic visual inspection of transparent materials has always been a challenging issue in industry due to complicated interference from reflection and refraction. In this paper, we present a study of machine vision system for automatic online inspection of transparent label defect on curved glass bottle. An area-array camera and a custom-made blue dome illumination device are introduced to capture high quality standstill image by eliminating reflection. To overcome the distortion issue on curved geometry shape, we have introduced the deformable template matching method for accurate location. An adaptive threshold selection strategy is proposed to effectively detect small scratch by using global and local threshold values together with Gaussian fitting algorithm. Considering the golden edge printing error, skeleton extraction and distance transformation are applied to detect the whole edge contour of Chinese characters with special font. Our visual inspection system has been deployed in a glass bottle manufacturing plant for on-line quality control. Field test result demonstrates that the detection accuracy reaches 99.5% at a speed of 60 pc/min for over 60,000 bottles.