A sequential machine vision procedure for assessing paper impurities

Highlights

• We present an innovative machine vision procedure for detecting particles and impurities in paper.

• A preliminary step separates defective paper patches from the non-defective ones.

• Defective patches are further analysed to assess the extension of the impurities.

• Combination of the two steps is highly effective providing over 96% accuracy.

Abstract

We present a sequential, two-step procedure based on machine vision for detecting and characterizing impurities in paper. The method is based on a preliminary classification step to differentiate defective paper patches (i.e., with impurities) from non-defective ones (i.e., with no impurities), followed by a thresholding step to separate the impurities from the background. This approach permits to avoid the artifacts which occur when thresholding is applied to paper samples that contain no impurities. We discuss and compare different solutions and methods to implement the procedure and experimentally validate it on a datasets of 11 paper classes. The results show that a marked increase in detection accuracy can be obtained with the two-step procedure in comparison with thresholding alone.

Introduction

Paper may contain particles of various types. In most cases these represent defects and impurities that need to be avoided; in other cases they are purposefully inserted in the paper to give the final product a peculiar visual appearance. In either situation the papermaking industry is increasingly concerned with the development of quick and reliable systems to detect and characterize such inclusions automatically. The growing attention towards environmentally friendly production policies and the consequent rise in production of recycled paper [14] – intrinsically more prone to contain defects – has rendered this need more and more compelling. The detection and characterization of particles can also help to determine the source of impurities in the production process, which can be subsequently amended and eliminated. This may reduce the use of chemicals in the bleaching phase, with beneficial effects on the environment. When speaking of defects, specific international standards [1], [2] provide definitions and quantitative means to assess their extent and the quality of the paper. Otherwise, when particles are actually desirable features of the product, their control may be performed in compliance with internal norms of the companies.

In the last twenty years, automatic visual inspection has benefited from the steady development of machine vision, whose applications now embrace a wide range of very diverse industrial products, such as wood [10], [17], textile [9], natural stone [7], exterior car parts [13] as well as food and agricultural products [3], [18], to cite some. In the papermaking industry, applications of machine vision are not uncommon and have covered, thus far, many problems like curl estimation [38], printability analysis [24], control of stripes and holes [29], sorting of waste paper for recycling [34], recognition of paper manufacturer and lot for forensic comparison [4] and characterization of fibre properties [11], [21].

Among the various applications, the identification of impurities has received significant attention, since such defects greatly affect the quality of final products. Within this field, Torniainen et al. [39] described an apparatus to measure dirt points on wet and dry pulp sheets through transmitted light reporting accuracy from 75% to 90%. Likewise, Duarte et al. [12] proposed a system for dirt inspection on pulp and paper based on hiearchical image segmentation. Later on, Campoy et al. [8] presented ‘InsPulp-I’, an inspection system for the pulp industry. More recently, interesting results have been obtained within the project ‘PulpVision’ [33], the aim of which is to detect dirt particles in pulp and classify them into different categories (i.e., bark, shives, etc.).

The literature shows that the common strategy to attack the problem consists of a preliminary image thresholding step to separate whatever kind of contraries from the background, followed by further analysis to classify them into one of some predefined categories. For such a strategy to work correctly, one has to implicitly assume that the paper patch under control does actually contain some type of particles; otherwise, if there are no particles at all, any image thresholding procedure is bound to produce unpredictable results, as we show in Fig. 1. To solve this problem, we propose a novel approach in which we first separate paper areas into defective and non-defective, then proceed to further analyse only the defective ones. Experimentally, we show that the method can provide an average increase in detection accuracy of about 25%.

In the remainder of the paper we first give a general overview of the method (Section 2), followed by a description of the materials and image acquisition devices used in the study (Section 3). In Section 4 we present and compare different solutions to implement the two steps of the method. The experimental activity is detailed in (Section 5), followed by the results (Section 6) and concluding considerations (Section 7). For the purpose of reproducible research, all the data and functions used in this study are available to the public [32].