Algorithm For Segmenting Overlapped Blisters of Coatings in Images Similar To Human Perception in Quality Control of Coatings

Appearance inspection of products as a quality control procedure is related to aesthetic faults that are subjective attributes. investigation of coating surface for defects done mostly manually is an example of this process. circular defects such as blisters, bubbles, craters, and pinholes are common defects occurring on coatings. however, machine vision systems proposed in most researches could not recognize them with a rate of more than 85%. low contrast between these defects and the background, their variation in size and shape, and overlapping one another are some factors. inspectors can easily distinguish the two or more blisters that overlapped each other and assess the size degree of blistering depending on the size of one blister instead of the whole combination. therefore, the overlap of blisters must be considered in the image processing algorithm to evaluate the degree of blistering correctly. for this approach, samples of paint with blistering defects of various sizes and frequencies were prepared. after the image acquisition of samples and preprocessing of images, blisters got extracted. watershed transform was applied on the resulted binary image of blisters to grant segmentation of overlapped ones. however, it resulted in some false segmentations. hence, an improvement in the algorithm combined invalid segments while not affecting the true ones. the proposed algorithm in this paper provided sufficient precision of 93% for assessing the size of blister defects by comparison with five inspectors’ results.

Algorithm for segmenting overlapped blisters of coatings in images similar to human perception in quality control of coatings

Appearance inspection of products as a quality control procedure is related to aesthetic faults that are subjective attributes. Investigation of coating surface for defects done mostly manually is an example of this process. Circular defects such as blisters, bubbles, craters, and pinholes are common defects occurring on coatings. However, machine vision systems proposed in most researches could not recognize them with a rate of more than 85%. Low contrast between these defects and the background, their variation in size and shape, and overlapping one another are some factors. Inspectors can easily distinguish the two or more blisters that overlapped each other and assess the size degree of blistering depending on the size of one blister instead of the whole combination. Therefore, the overlap of blisters must be considered in the image processing algorithm to evaluate the degree of blistering correctly. For this approach, samples of paint with blistering defects of various sizes and frequencies were prepared. After the image acquisition of samples and preprocessing of images, blisters got extracted. Watershed transform was applied on the resulted binary image of blisters to grant segmentation of overlapped ones. However, it resulted in some false segmentations. Hence, an improvement in the algorithm combined invalid segments while not affecting the true ones. The proposed algorithm in this paper provided sufficient precision of 93% for assessing the size of blister defects by comparison with five inspectors’ results.