Paper sketch-based 3D Model Retrieval by Viewpoint entropy-based Adaptive View Clustering

Title: 3D model Retrieval with adjustable view clustering based on viewpoint entropy

Bo Li,yijuan Lu,henry Johan

Abstract: Searching for 3D models based on freehand sketches is intuitive and important for many applications, such as sketching-based 3D modeling and recognition. We propose a sketch-based 3D model retrieval by using the adjustable view clustering and shape content matching based on the viewpoint entropy. Different models consist of different visual complexities, so there is no need to fix the same number of representations for each model. For this reason, we propose to measure the visual complexity of the 3D model by using the distribution of the sampled viewshed and the complexity values, and we can dynamically determine the number of views represented. Finally, we use fuzzy C-means to cluster sampling view based on the viewpoint entropy value. We tested our algorithms on the latest two 3D model retrieval databases and compared them with the other four advanced methods. The results show that our algorithm has good performance.

1.introduction:

Using sketches to retrieve related 3D models is an intuitive and easy way for users. It is useful for modeling scenarios or for identifying similar applications.

Currently, some sketch-based 3D model retrieval algorithms are presented. However, most of these methods are compared with the 3D model pre-defined sample view and 2D sketch. However, these sampled policies do not guarantee that the sampled views extracted are sufficient to depict the 3D model because they do not consider different complexities of the models. In fact, there is no need to compare 13 or 14 views with a simple model such as a sphere or a cube, and more views need to be sampled from a complex model. This means that we need a sampling strategy that can be adjusted.

Based on the above findings, we propose that the visual complexity of the 3D model (based on the measurement of the shape complexity of the visual information) determine the different number of representation views. This paper presents a novel 3D visual complexity measurement based on the uniform sampling view of 3D model for visual entropy distribution. Furthermore, the fuzzy C-means method is used to determine the number of cluster centers with the value of the viewpoint entropy. The shape context matching algorithm is then used to match the 2D sketch and the various representations of the 3D model between the views. Our algorithm tests were proven to be effective on two of the most recent 3D model retrieval libraries.

The main contribution of this paper is the following three aspects:

<1> quantitatively studied the complexity of 3D models. Based on this, we propose an effective measure of 3D visual complexity by measuring the entropy of information-related viewpoints.

<2> We propose methods that can be used to adjust the number of representations of each 3D model to determine the representation of a view.

<3> Our research leads the way in the research of this problem.

2 related work

2.1 Sketch-based 3D model retrieval

Based on different sampling strategies, sketch-based 3D model retrieval techniques can be divided into two categories: matching sketches from predefined fixed viewpoints, and clustering to get a view to match.

With a predefined view. As mentioned above, most of the existing sketch-based 3D model retrieval algorithms are compared by a series of views that are obtained in a predefined direction. "Work done by predecessors"

The disadvantage of these strategies is that their presentation models are simply represented from the selected view. This also drives us to develop a sketch-based retrieval algorithm for dynamically clustered sampling views.

Use the cluster view. Compared to the method based on predefined views, there is less research on the strategy based on view clustering. "Work done by predecessors"

2.2 Complexity of shapes

The method of geometrical complexity is rossignac from five aspects: Algebra, topology, morphology, composition and representation. At present, a new trend is to measure the visual complexity of 3D models. This is also the basis for computer vision and 3-dimensional human perception: a 3D object can be viewed as a collection of 2D views. Using information theory to measure the complexity of 3D models is also consistent with human perception.

It is considered an effective method to characterize the visual information characteristics of a sample view of a 3D model using information-related measures, so it is useful to measure 3D shape complexity. Based on this, Vazquez presents the viewpoint entropy to depict the number of information contained in a view, and in this way, they developed a method for automatically finding a collection of views with a large viewpoint entropy value.

3 Viewpoint entropy distribution-based on view clustering

In this paper, we propose a measurement of 3D visual complexity based on a series of views of the entropy distribution of viewpoints. We then use the 3D visual complexity metric to our sketch-based 3D model retrieval algorithm to determine the number of 3D model representation views (i.e., clustering centers). Finally, based on the value of the viewpoint entropy of the sampled view, the Fuzzy C-means algorithm is used to select the number of views represented for each model.

The entropy distribution of the viewpoint. We use the loop decomposition algorithm to decompose a unit positive 20 polygon (represented by L0) n times, and the resulting shape is represented by Ln. All 3D models are first scaled to the unit ball and then drawn in 3D to obtain his orthogonal projection. We use the method of viewpoint entropy calculation in citation [TFTN05]. That is, for a 3D model with M faces, the viewpoint entropy of one of his views is defined as follows.

　　　　　　　　　　　　　　　　

Where AJ is the visual projection area of the J-Plane of the 3D model, A0 is the background area. S is the total area of the window, and the model is based on

To calculate the total area.

Paper sketch-based 3D Model Retrieval by Viewpoint entropy-based Adaptive View Clustering