3D Object Retrieval Based on Similarity Calculation in

2017 IEEE/ACS 14th International Conference on Computer Systems and Applications

3D object retrieval based on similarity calculation in 3D Computer Aided Design systems Ahmed Fradia, Borhen Louhichib , Mohamed Ali Mahjouba, Benoit Eynardc a

LATIS laboratory of Advanced Technology and Intelligent Systems, University of Sousse, Sousse 4023, Tunisia b LMS, ENISo, University of Sousse, Sousse 4023, Tunisia c Sorbonne universités, Université de Technologie de Compiègne, CNRS, UM R 7337 Roberval, Compiègne, France

[email protected]; [email protected]; [email protected]; [email protected] Abstract—Nowadays, recent technological advances in the acquisition, modeling and processing of three-dimensional (3D) objects data lead to the creation of models stored in huge databases, which are used in various domains such as computer vision, augmented reality, game industry, medicine, CAD (Computer-aided design), 3D printing etc. On the other hand, the industry is currently benefiting from powerful modeling tools enabling designers to easily and quickly produce 3D models. The great ease of acquisition and modeling of 3D objects make possible to create large 3D models databases, then, it becomes difficult to navigate them. Therefore, the indexing of 3D objects appears as a necessary and promising solution to manage this type of data, to extract model information, retrieve an existing model or calculate similarity between 3D objects. The objective of the proposed research is to develop a framework allowing easy and fast access to 3D objects in a CAD models database with specific indexing algorithm to find objects similar to a reference model. Our main objectives are to study existing methods of 3D objects similarity calculation (essentially shape-based methods) by specifying the characteristics of each method as well as the difference between them, and then we will propose a new approach for indexing and comparing 3D models, which is suitable for our case study and which is based on some studied previously methods. Our proposed approach is finally illustrated by an implementation, and evaluated in a professional context.

hence the similarity calculation measures will be different and various algorithms are proposed to allow searching in 3D model databases. Therefore, a method cannot be effective in all situations, and cannot offer multiple search possibilities, within many search fields, otherwise, the algorithm complexity will increase and it will not minimize response time of search queries, especially in the current context of vast 3D databases containing thousands (sometimes millions) of 3D models. This paper presents a new approach for description, indexation and comparison of 3D objects from CAD models, for the purpose of establishing a model retrieval solution for designers. It is composed by five sections. In the first section, we aim to identify the context of our research project, presenting the importance of 3D modeling in the modern industry / design, as well as a general presentation of the problem of 3D CAD model retrieval, in order to allow an optimal view of the problem that we seek to solve in our work. Next, a second section aimed at putting the finger on the state of the art of different notions in relation to our subject, in particular, the 3D object content-based retrieval approach, especially shape-based methods. The third section of our paper is devoted entirely to the presentation of our proposal; therefore, it contains the description necessary for understanding the system, accompanied by the necessary scenarios and diagrams. Then, a fourth section to validate our proposal, details the realization of a prototype implemented on a case study. Finally, our approach will be evaluated in the last section.

Keywords-Image retrieval; Reference model; CAD; 3D object;

I. INT RODUCTION The use of 3D models is currently growing rapidly. Especially with the rise, the diversity and the simplicity of 3D modeling software, then, it becomes increasingly easy to create a very realistic and detailed 3D model. Therefore, some questions that arise are interested in the best exploitation of the huge collections of existing 3D models, what is the right way to research in these collections as efficiently and as simply as possible? And how to find a 3D model already designed, and consequently save the design time and cost? To answer these questions, many studies have been interested in indexing 3D objects. Accordingly, powerful search methods have been implemented based on a specific indexing process, in order to describe a 3D model in a very synthetic manner according to well selected characteristics. On the other hand, the description of a 3D model is too strongly dependent on the associated domain, as well as the type of the model queried and the modeling way, so an indexing method cannot necessarily describe an object in the same way than another,

2161-5330/17 $31.00 © 2017 IEEE DOI 10.1109/AICCSA.2017.101

II.

CONT EXT

The field of 3D vision evolves very quickly, thanks to its attractive side and the appearance of new 3D modeling technologies. Since 3D modeling is used to make objects in 3D, it is useful in many sectors, it can thus be used in the automotive industry, in architecture to draw plans, in science to present phenomena in a virtual way, it is also very useful in the field of teaching, video games and animated films… A. 3D CAD Modeling explosion in design and manufacturing systems Computer-aided design (CAD) software allows designing scenes or objects in 3D, exploring design ideas, visualizing concepts, and simulating design performance in a real-world context according to Robertson and Radcliffe [1], and Leng and Vonderembse [2]. CAD tools can help engineers and designers to understand their 3D designs virtually before 160

manufacturing them, by connecting each phase of the design process through a single digital model, they also test and optimize designs, which can promote innovation, improve quality, reduce costs and accelerate manufacturing, according to Leng and Vonderembse [2]. Today, the use of CAD software is widespread. They evolve every day, and they are used not only in aeronautics, mechanical, architectural and automotive sectors but also in education, medical and food industry, that is what we find in many works like Ye et al. [3], and Daud et al. [4]. In conclusion, the use of a CAD tool has widened considerably over the last two decades in all areas. Besides its use in its primary role based on virtual modeling, its evolution and ease of handling has allowed it to open up to a secondary function where aesthetics and optimization play a key role.

terms of modeling way. As a result, we can distinguish several generations of CAD systems that can be classified from a historical point of view according to Tornincasa et al. [11], Figure 1. Computer-aided Drafting was the first CAD generation; the objects were presented in a two-dimensional plane. After that, the CAD has passed to other generations based on the 3D geometric modeling that is defined as the numerical representation of a space object. Up to the parametric CAD of nowadays presented by Bodein et al. [9]. There are several approaches of 3D geometric modeling: wireframe modeling, surface modeling, and volume modeling. B. 3D object retrieval: stat of the art As mentioned in the first section, 3D objects are becoming more and more numerous and their massive storage requires the creation of specialized databases. It is for this reason that

B. 3D CAD models retrieval as a real challenge The volume of 3D CAD models existing in the web is constantly increasing. Also, industries using CAD tools constitute, internally, their own databases which generally can contain several thousand 3D models. As a result, management, archiving and exploitation of these files can become increasingly difficult, and being able to browse such databases to find a 3D model can be considered as daunting task in view of their large size, as well as the specific format of 3D data. According to Tao et al. [5], "In industries, enterprises have suffered from tremendous waste due to the repeated design phenomenon. Obviously, effective information retrieval tools are helpful for solving this problem." On the other hand, the use of an existing model as a basis for designing a new model exceeds 75% of the newly designed models adds David G. Ullman [6]. So, according to Bai et al. [7] information retrieval makes possible to adapt a model instead of redesigning a new one. As a result, the challenge facing the industry today is how to take advantage of one of its greatest assets, its CAD files, and how to locate 3D model/parts efficiently so that it can be reused every time is necessary, Gunn [8]. And therefore be more productive by searching for existing files, models or parts to avoid the cost of redesigning, saving time and consequently accelerating design processes according to Bodein et al. [9].

Figure 1. T he historical improvements of the CAD modeling tools

industries, designers and scientific laboratories constitute, internally, their own 3D model libraries which usually contain several thousands, sometimes millions, of 3D objects. Being able to browse such databases can be a tedious task once their size becomes large and their types become heterogeneous (use of different 3D design tools). Indeed, the major asset is quicker and reliable access to documents or to databases to find an existing model. In order to effectively address this problem, many search techniques have been put in place to offer multiple interrogation capabilities. Many search engines offer a text-based query means, often reduced to keywords only. This method is very suitable for the text document where the information is contained in a set of sentences, which it is generally possible to synthesize in a few keywords. But, when the documents contain visual information such as photos, videos or 3D objects, these search techniques show a certain weakness. Other criteria have been added to facilitate the search task, such as creation date, version, designer information, etc. These techniques offer a semantic search based on the 3D object metadata, but they have the same drawbacks as text-based search in terms of lack of accuracy and indexing efficiency. To overcome this

III. 3D OBJECT RET RIEVAL : METHODS AND RELATED WORKS The 3D models are represented by different manners according to the needs, and each modeling software proposes its own representation way. Consequently, for an algorithm of 3D files indexation, the description of a 3D model can be differentiated according to the field of application and especially according to the modeling way, because the objective of each indexation method is to characterize 3D objects by relying on information about their structures. In the same context, the CAD objects, by their particular nature, and their topological properties, make possible to implemen t indexing methods that are specific to them. A. 3D modeling methods in CAD systems CAD software uses geometric modeling techniques to model an object, Agoston [10]. Since its first appearance in the 1950s, CAD technology has realized many evolutions in

161

difficulty, in order to quickly obtain a 3D model and basing the indexing on the visual aspect of the document. The new indexing processes are based on the content, comparing the shape of each 3D object with the user's query.

used in 3D indexing, Novotni and Klein [20], Lakehal and El Beqqali [21], Zhang et al. [22], Zhao et al. [23]. The representation of the 3D object by a transform is generally defined by the projection of the function characterizing the 3D object (surface or volume) on a family of characteristic functions. Also, we can talk about other approaches that are based on product information, they consist in assigning a unique descriptor (code or signature) for each 3D object, these methods are inspired by the manufacturing industry, they consist in grouping parts by families through the use of a code describes shape characteristics and dimensions, the code also describes techniques for fabrication in order to collect parts having similar manufacturing circuits. This classification technique is called GT (Group Technology), and among the most commonly used methods we can cite Opitz code used in many works such as Zehtaban et al. [24]. Although GT approaches are used in the manufacturing industry, they can be very useful in the 3D objects search, so we distinguish many GT code-based research methods.

C. 3D shape-based searching Content-based searching seems to be a promising solution for indexing and browsing 3D object databases, there is interest in describing the shape of 3D models. Thanks to this method that an efficient comparison, by means of an adapted similarity measure, will make it possible to find objects stored in databases which are similar to the request provided by the user. The representation of 3D shapes can be envisaged in several ways, according to what we wish to describe and which comparison criteria to apply, in this context, there are several families of methods allowing shape representation. So, analysis of the literature shows that there are a variety of 3D shape representations that derive from different aspects of 3D objects. Some of them rely directly on the geometric and topological properties of the 3D object, while others rely on the 2D projections of the 3D object. In this context, we have chosen to classify briefly existing methods in families of approaches based on several works such as Iyer et al. [12], Hong et al. [13] and Li et al. [14]. Statistical approaches consist in characterizing 3D objects by one or more distributions of mathematical descriptors of shape. These distributions are usually represented in the form of histograms. The comparison of objects is done by calculating a distance between histograms. Shape representations based on these approaches can be classified into two main categories: local representations, which rely on the local characteristics and using mainly the calculated curvature at any point of the 3D object like works proposed by Alaoui Mhamdi and Ziou [15], and Koenderink and van Doorn [16], and global representations, which rely on a shape function that measures the geometric properties of the object in its entirety by calculating for example distances between points on the surface of the object, or 3D moments, such works proposed by Osada et al. [17], and Adan et al. [18]. Another family is that reunites the structural approaches, the objective of these methods is to characterize 3D objects, relying on information about high levels on their structures. In the past few years, graphs have been used to represent the structure of an object. This type of representation makes it possible to represent the variations and the structural differences between objects and to combine them with graph matching algorithms, as example the work proposed by Barra and Biasotti [19]. So, the comparison of the models is done by comparing their graphs. During the indexing phase, these methods extract a structural representation of the following 3D object according to different methods, based on topological properties (CAD approaches), based on volumetric decomposition, or based on skeletons, according to Li et al. [14]. Transform-based representations have often been proposed in the literature as descriptors of 2D shape. We can cite: The Fourier descriptor, the descriptor of Hough, the geometrical moments and the moments of Zernike. By 2D-3D analogy, most of these transforms have been extended and

IV. A FRAMEWORK FOR 3D CAD OBJECT RET RIEVAL BASED ON SIMILARIT Y CALCULATION

Figure 2. T he structure of the five first digits of Opitz code

Numerous solutions for 3D objects indexing based on the previously seen approaches have emerged in recent years, driven by the massive appearance of 3D models. Despite their increasingly improved search efficiency, however, current solutions still remain either incomplete or too costly in computing time and therefore do not adequately meet the user's needs. The bibliographic work allowed us also to observe that the efficiency of a 3D objects searching method can be estimated according to several factors such as the precision, the speed, as well as the complexity of the algorithm. It always depends very much on the context and the intended application, according to Li et al. [14]. Therefore, research works are still ongoing in order to optimize an existing approach, or to propose a new more efficient approach for a specific field.

162

This section introduces our contribution in the field of 3D object search. We will propose a complete search process based on the similarity calculation, we will apply our approach in the field of 3D CAD, to find 3D parts.

defined by Cha [27] to generate an index between two Opitz codes presented as two vectors (P and Q), in the same way we have chosen to adapt the same process to calculate the similarity. ∑ ( ∗ ) (1) ∶ (P, Q) =

A. Proposal approach for 3D model description Indexing is a method of codifying the content o f a document in order to speed up the querying of large databases. This operation consists in analyzing the different properties of the 3D model in order to extract the main characteristics. This description is a synthetic, and relevant representation of the content. The main task is therefore to select the important data present in the object and to codify them in a description adapted to the extracted information. Finally, it is possible to quickly compare two objects by analyzing their description. 3D CAD objects by their particular nature and in particular by their topological properties allow to implement their own description processes according to Wei Sun et al. [25]. As a result, shape description approaches for CAD objects were mainly based on geometry and topology data because the easy retrieval of this type of information from adapted modeling methodologies in CAD tools or exchange formats between them such as STEP standards. In our proposed approach, we will also exploit geometric and topological information by assigning an alphanumeric code for each 3D object. Our description is inspired by Opitz code which has been used frequently in industry to classify parts, this type of coding offers an extensive description capability considering his hybrid (mixed) nature of coding explained by Opitz h. [26]. On the other hand, it offers a classification simplicity during comparison process, for example, the distinction between rotational and no rotational parts is made once the first digit of descriptor is compared (Figure 2). Since our approach is multi-criteria and is not limited to the geometric and topological description, we have chosen to offer other comparison metrics to optimize the search process, by exploiting other information provided by the 3D mo del. Thereby, an extension was added to the descriptor summarizing other extracted attributes.

Concerning the second descriptor code block which follows the geometric and topological representation (5 first Opitz code digits), and which groups other 3D shape features, we have assigned a weight to each active digit (which has a value). The activation mechanism of comparison criteria will be shown in the next section which will present application graphical interfaces. So the similarity index between two 3D objects O1 and O2 will be given finally by the followin g formula (2). (2) :

(

,

) = S(

,

)∗

+

(

(

,

)∗

)

Ic: The similarity index of each criteria c. n: Similarity criteria number. ws : The weight assigned to the index calculated by (1). wc: The weight assigned to the similarity index of each criteria (Ic). With the following conditions: + ∑ ( ) = 100% 0≤ ( , )≤1 0≤ ( , )≤1 The weight (ws ) assigned to the index calculated by (1) and which reflects the geometric and topological similarity takes 100% in the absence of the other criteria, otherwise it takes the value 80%, and the remaining 20% is allocated to others comparison criteria in a dynamic way, depending on the user's setting before the launch of the request. In this way, the formula (2) will always give an index between 0 and 1, and by a simple multiplication, we can render the result in the form of a percentage of similarity. For example, if the user activates the search criterion Materials and Dimension (Figure 4), these two criteria will be taken into consideration. Then, = 80% = = 10%, hence a model identical to and the query in form and dimension and different in terms of construction materials will obtain a similarity index of 90%.

B. Search process and similarity index calculation The purpose of the user is to find objects that have a shape similar to his query. Since we are in possession of the shape description of each 3D object, the query and the database object, we can compare them. For this purpose, we define a similarity measure allowing to give an index of resemblance between two descriptors. Here we must briefly remind the utility of search by a 3D model as input. Indeed, in the majority of cases, designers need to design or make a small change on a model already designed. In this case, the textbased search is limited, and a shape-based search will be too useful, and specify the search criteria will optimize the desired result. Then the designer will enter a model Similar to the one sought, then, the model entered may be an old one similar in shape, or a model newly and rapidly designed by the user belonging to the same family of desired model. To evaluate the distance between two Opitz codes previous works have applied a cosine coefficient method like Zehtaban et al. [24], which has proposed the equation (1)

C. Framework Architecture As mentioned above, for our present paper, we will propose a framework to find a 3D CAD object according to a user query, based on the similarity calculation through a shape descriptor. As illustrated in Figure 3, the main idea is (1) to enter a 3D object and to launch a parameterized request, (2) to extract a shape descriptor from the 3D query model, and to give the ability to calculate similarity with database stored models or not stored models (model in raw format) (3) to provide ergonomic user interfaces which allow user to make

163

Figure 3. Global view of the search process from the description of the user request to its comparison with the database of 3D objects via the human-machine interface allowing the dialogue between the user and the search engine.

multi-parameter request and to display results. When the user wishes to search for a specific 3D object, he must formulate his search by means of a request, this request is in the form of a 3D object. Once the user's request has been obtained, it is necessary, in order to compare it, to describe its s hape by assigning it a signature (the indexation process). As a second step, it is necessary to apply the similarity measure defined previously between two synthetic descriptions, as well as a procedure allowing the comparison of the query to the entire database which already contains stored signatures (offline indexation), or with raw-formatted models (a 3D object, exchange format STEP...), so, our approach offers the possibility of comparing with 3D objects not previously stored in the database by indexing them online. Finally, to return to the user a list of relevant 3D models, it is enough to filter the result according to initial request parameters. Thus, it will be possible to return as a result the list of similar objects ordered according to their distance to the request (Figure 5).

V.

RESULT S, EVALUAT ION AND DISCUSSION

As we can see in the result shown in Figure 6, the search process for the 3D model example returns seven most similar models; The first model, which has 92% as an index of similarity, is almost identical to the given model, the only difference is the material (the request is made of iron, but the result is made of wood). So we can conclude that the weight assigned in the user's search query to the Materials criterion is 8%. Given the specificity of our proposed approach, which is concerned with a specific 3D object type (3D CAD objects), to evaluate our approach, we invited a team of CAD 3D designers within a mechanical research laboratory to try our solution during their work. The team composed by 15 researchers shows a very good feedback. All the evaluations

D. Application user graphic interfaces In order to understand the functioning of our research process approach, we will briefly explain it through user interfaces. When the user wishes to search a specific 3D object in the database, he forms his search using a parameterized query as shown in Figure 4. Then he has the ability to set the search parameters. It is offered to the user also to s pecify the directory that contains the files to be compared out of the database (to index them online). Finally, the result of the search will be returned in the form of a graphical interface which lists the most similar parts to the user query as shown in Figure 5. Figure 4. T he search parameters user interface

164

Figure.5. T he result user interface

Figure.6. Example of 3D object searching through our framework [8]

carried out show the good performance of our approach in terms of result relevance, as well as the simplicity provided in the reuse of the 3D models. But the only defect detected in the evaluation of our approach is the late response time of the application in case of a large number of files to be indexed online, (the 3D models in their raw format). VI.

[9] [10] [11]

CONCLUSION AND FUT URE WORK

[12]

Indexing 3D CAD objects was approached in a new way during this paper. First, we studied content-based approaches to indexing 3D objects and we proposed a new shape-based method. Our method assigns a signature by extracting quickly and precisely information characterizing the object shape and provides an interesting approach to online and offline index large 3D object databases by calculating the similarity according to a proposed formula. Our research presents different perspectives, on the one hand, we observed a reduction in performance when the number of models to index online increases, so it would be interesting to modify the metrics used during indexing. Also, an interesting research track, would be to index other types of 3D CAD files. On the other hand, future work could be to create a distributed web platform, querying databases of 3D CAD objects from different companies implementing our approach in a multi-CAD context using Big Data technologies which offers a huge storage and computing power.

[13] [14]

[15] [16] [17]

[18] [19]

REFERENCES [1] [2]

[3] [4]

[5]

[6] [7]

B. F. Robertson and D. F. Radcliffe, “Impact of CAD tools on creative problem solving in engineering design,” CAD Comput. Aided Des., vol. 41, no. 3, pp. 136–146, 2009. C. Leng and M. A. Vonderembse, “ Mediating effects of computeraided design usage : From concurrent engineering to product development performance,” vol. 24, pp. 494 –510, 2006. X. Ye, W. Peng, Z. Chen, and Y. Y. Cai, “T oday’s students, tomorrow’s engineers: An industrial perspective on CAD education,” CAD Comput. Aided Des., vol. 36, no. 14, pp. 1451–1460, 2004. M. F. Daud, J. M. T aib, and R. S. Shariffudin, “ Assessing Mechanical Engineering Undergraduates’ Conceptual Knowledge in T hree Dimensional Computer Aided Design (3D CAD),” 3rd Reg. Conf. Eng. Educ. Res. High. Educ., vol. 56, no. Ictlhe, p p. 1–11, 2010. S. T ao, Z. Huang, L. Ma, S. Guo, S. Wang, and Y. Xie, “Partial retrieval of CAD models based on local surface region decomposition,” CAD Comput. Aided Des., vol. 45, no. 11, pp. 1239–1252, 2013. David G. Ullman, T he Mechanical Design Process, 4th ed. 2010. J. Bai, S. Gao, W. T ang, Y. Liu, and S. Guo, “ Design reuse oriented partial retrieval of CAD models,” CAD Comput. Aided Des., vol. 42, no. 12, pp. 1069–1084, 2010.

[20] [21] [22] [23]

[24] [25] [26] [27]

165

G T . Gunn, The mechanization of design and manufacturing. Scientific American 1982;247(3):114–30. Y. Bodein, B. Rose, and E. Caillaud, “ A roadmap for parametric CAD efficiency in the automotive industry,” CAD Comput. Aided Des., vol. 45, no. 10, pp. 1198–1214, 2013. M. K. Agoston, Computer Graphics and Geometric Modeling: implemen tation & algorithms. 2005. S. T ornincasa, P. Torino, and P. Torino, “THE FUTURE AND THE EVOLUT ION OF CAD II-1 II-2,” no. September, pp. 11–18, 2010. N. Iyer, S. Jayanti, K. Lou, Y. Kalyanaraman, and K. Ramani, “Threedimensional shape searching: state-of-the-art review and future trends,” Comput. Des., vol. 37, no. 5, pp. 509–530, 2005. T . Hong, K. Lee, and S. Kim, “ Similarity comparison of mechanical parts to reuse existing designs,” CAD Comput. Aided Des., vol. 38, no. 9, pp. 973–984, 2006. B. Li, C. T ang, X. Zhu, X. Chen, Y. Su, and Y. Cai, “ A 3D shape retrieval method for orthogonal fringe projection based on a combination of variational image decomposition and variational mode decomposition,” Opt. Lasers Eng., vol. 86, pp. 345–355, 2016. M. A. Alaoui Mhamdi and D. Ziou, “A local approach for 3D object recognition through a set of size functions,” Image Vis. Comput., vol. 32, no. 12, pp. 1030–1044, 2014. J. J. Koenderink and A. J. van Doorn, “Surface shape and curvature scales,” Image Vis. Comput., vol. 10, no. 8, pp. 557–564, 1992. R. Osada, T . Funkhouser, B. Chazelle, and D. Dobkin, “ Matching 3D models with shape distributions,” Proc. - Int. Conf. Shape Model. Appl. SMI 2001, pp. 154–166, 2001. A. Adan, C. Cerrada, and V. Feliu, “Global shape invariants: A solution for 3D free-form object discrimination/identification problem,” Pattern Recognit., vol. 34, no. 7, pp. 1331–1348, 2001. V. Barra and S. Biasotti, “3D shape retrieval using Kernels on Extended Reeb Graphs,” Pattern Recognit., vol. 46, no. 11, pp. 2985– 2999, 2013. M. Novotni and R. Klein, “Shape retrieval using 3D Zernike descriptors,” Comput. Des., vol. 36, no. 11, pp. 1047–1062, 2004. A. Lakehal and O. El Beqqali, “ New Method for 3D Shape Retrieval,” Int. J. Comput. Sci. Inf. T echnol., vol. 3, no. 5, pp. 89–98, 2011. Y. Zhang, F. Jiang, S. Rho, S. Liu, D. Zhao, and R. Ji, “3D object retrieval with multi-feature collaboration and bipartite graph matching,” Neurocomputing, vol. 195, pp. 40–49, 2016. S. Zhao, H. Yao, Y. Zhang, Y. Wang, and S. Liu, “View-based 3D object retrieval via multi-modal graph learning,” Signal Processing, vol. 112, pp. 110–118, 2015. L. Zehtaban, O. Elazhary, and D. Roller, “A Framework for Similarity Recognition of CAD Models,” J. Comput. Des. Eng., vol. 3, no. 3, pp. 274–285, 2016. Wei Sun, Ma T ie-Qiang, and Guo Li, “A New CAD Models Retrieval Method Based on Shape Similarity.” Information Technology Journal, 2009. Opitz H. A Classification System to Describe Workpieces. T ranslated by Acton T aylor. New York: Pergamon Press; 1970. S. Cha. Comprehensive Survey on Distance / Similarity Measures between Probability Density Functions. Int. J. Math. Model. Methods Appl. Sci, 4(1):300–307, 2007.