تجزیه و تحلیل عملکرد از توصیف رنگ برای مرتب سازی پارکت

In this paper we consider the problem of colour-based sorting hardwood parquet slabs into lots of similar visual appearance. As a basis for the development of an expert system to perform this task, we experimentally investigate and compare the performance of various colour descriptors (i.e.: soft descriptors, percentiles, marginal histograms and 3D histogram), and colour spaces (i.e.: RGB, HSV and CIE Lab). The results show that simple and compact colour descriptors, such as the mean of each colour channel, are as accurate as more complicated features. Likewise, we found no statistically significant difference in the accuracy attainable through the colour spaces considered in the paper. Our experiments also show that most methods are fast enough for real-time processing. The results suggest the use of simple statistical descriptors along with RGB data as the best practice to approach the problem. Highlights ► We present a comparative performance analysis of colour descriptors and spaces for parquet sorting. ► We found no statistically significant difference in the accuracy of the colour descriptors and spaces considered in the experiments. ► Fast and accurate parquet sorting can be obtained through simple statistical descriptors in the RGB colour space.

Wood is a widely used and greatly appreciated material. Countless are its applications in various industrial sectors including construction, interiors, furniture and shipbuilding. Like other products such as natural stone, ceramics, leather and similar, wood is mostly appreciated for its appearance; a feature that determines, to a great extent, its price. When wood is used for flooring, decking or façade cladding (in this case we usually refer to it as engineered wood), strict selection procedures are needed to assure satisfactory aesthetic results. To obtain beautiful and uniform surfaces, wood has to be carefully graded by fibre type and colour tone. In an increasingly globalised and competitive market, it is mandatory that wood products – particularly those of high range – be virtually extent of any defects. In an endeavour to meet such requirements and increase market shares, producers are trying to drastically improve their quality standards. In this context quality inspection plays a central role. As noted by Bombardier and Schmitt (2010), wood quality inspection involves two different and clearly separated problems: (1) detection, localization and classification of surface defects; and (2) sorting products into lots of similar appearance. In the parquet industry the two processes are usually carried out sequentially and in this order. Both can be performed either manually or automatically. Technically speaking the first problem is referred to as grading and is related to detecting, measuring and counting superficial defects like knots, pockets, stains, veins, cracks, etc. Domain-specific standards ( DIN-1611, 2002, DIN-EN-975-1, 2011 and DIN-EN-975-2, 2004) define different wood grades on the basis of the number and size of such defects along with procedures to their measurement and detection. As for the second problem, we can find it referred to as sorting ( Lu, Conners, Kline, & Araman, 1997), colour classification ( Kurdthongmee, 2008), or, again, grading ( Faria et al., 2008 and Vienonen et al., 2002). To avoid confusion, throughout this paper we use the term grading to refer to the first problem and sorting for the second. When performed manually, grading is stressful and time consuming, though, in general, not particularly demanding, since defects are usually quite evident. In contrast, sorting products into groups of similar appearance is more subtle, since products of the same class may have differences in tone which can be very slight and difficult to detect even to a trained eye. In addition this process requires more than one slab to be observed at the same time. Subjective and environmental conditions, tiredness, boredom and other factors can significantly affect the outcome of the process. To this we should add that recent studies showed how colour perception can be significantly influenced by age and socio-economical level of the subject (Kose, 2008). Our personal experience indeed confirms that different operators can produce very dissimilar results. Beginning with these considerations, it is therefore no surprise that the agreement between different operators can be as low as 60% (Roz˘man, Brezak, & Petrovic, 2006). As a consequence, manual inspection procedures can produce batches of products with significant variations of the visual appearance, causing sales returns and significant economical losses. Parquet producers are therefore more and more concerned with the development of computer vision systems capable of carrying out automatic quality control procedures. In this paper, in particular, we are concerned with the second problem, that of measuring and comparing the visual appearance of parquet slabs in order to sort them according to suitable similarity criteria. More specifically we focus on the problem of colour sorting slabs of the same grade and quality. This consists of dividing a previously graded batch of parquet hardwood into different colour tones, among which differences in colour are usually very slight, yet noticeable. To this end we experimentally compare the effectiveness of a set of colour descriptors and spaces. We also discuss issues related to image acquisition and processing including computational time, which are of primary importance when it comes to designing and implementing practical, real-time solutions. The remainder of the paper begins with a brief survey of related research (Section 2). In the following sections we give a description of the materials (Section 3) and methods (Section 4) used in our research. In Section 5 we present the experimental activity followed by results (Section 6) and conclusions (Section 7).

In this paper we have presented a performance analysis of different colour descriptors and spaces for sorting parquet slabs into classes of similar visual appearance. To the best of our knowledge, this is the first comprehensive study on the subject based on statistical analysis and reproducible experiments. The overall results show, on average, a low error classification process with about 90% accuracy. Most methods are also computationally inexpensive, therefore suitable for real-time processing. This outcome is satisfactory, considering that the results have been obtained with standard industrial equipment – specifically a single-sensor camera – and a very simple classifier (1-NN). We believe that the overall accuracy could be significantly increased by adopting higher-level imaging devices (i.e.: 3-sensor camera) and more sophisticated classifiers. The comparative analysis showed no significant difference between the descriptors and colour spaces considered in the paper, therefore suggesting – from both standpoints of accuracy and computational efficiency – the use of simple statistical descriptors along with RGB data as the best practice.