A robust algorithm for detecting pinholes in transparent plastic films

摘要

Study of the correlation between the mechanical properties of plastic films and the pinhole propagation in them is essential for the manufacture of plastic films with desired properties. Accurate information on the size and number of pinholes is required to facilitate this study. A computer vision based technique which automatically detects pinholes in microscopic images of transparent plastic films and also provides accurate and consistent information about the size, number and location of pinholes is presented. Pinholes appear as white blobs with a very dark boundary in the microscopic images of transparent plastic films. The microscopic image is first mapped into three different colors. The boundary pixels of each white blob in a binarized image are then extracted. The algorithm then searches the neighbors of the boundary pixels in the mapped image to verify a very important property of pinholes. Each blob is then classified as a pinhole or not based on the comparison of some quantified features to critical values. Experimental runs of this algorithm have given very consistent and impressive results.