Feature Belts-based3D Shape Retrieving

In recent years, sketch-based3D shape retrieving is becoming more and morepopular. A3D shape retrieval system based on sketch of feature lines are proposed in[Sketch-based3d Shape Retrieving2012]. Users can get the3D model what they wantaccording to their own hand drawn sketches of the model.Compared to the image,sketch has become an important process for people to express their own ideas.Sketchis simple and convenient to input, also can express the shape information of3d modelintuitively. However, the disadvantages of sketch are significant, sketch is formed bysome feature lines simply.Therefore, the amount of shape information contained insketch is special few, which led to the features of sketch is not obvious and not easy tocalculate. Therefore, we put forward some methods to improve the shortcomings ofthe lack of information and improve the process of sketch feature extraction. Based onthe improvement of sketch, we propose a3D shape retrieval system based on thesketch of feature belts. Compared to the existing retrieval system, we do someimprovement on how to extract the feature lines better and how to access a betterfeature vector of sketch.Firstly, we put forward the feature belts algorithm to replace the feature lines.Compared to the feature lines, feature belts has the advantages of reduce the quantityif calculation and contains more shape information of3D model in two aspects.We adopt the local polynomial instead of the global surface to reduce thecomputational scale in estimating discrete curvature and computing partial derivatives.Then we compare the curvature of the point with its neighborhood to determine theextreme points of curvature.We use the curves with width called feature leaves instead of lines to connectthe extreme point. Meanwhile we hope that the curves have different width at thepoints of different curvature on the mesh, by which the human can perceivethree-dimensional structure better from feature leaves. At the beginning, we thoughtthat the width is proportionate to the principal curvature of the points. However, according to the human’s visual perception, the width the feature leaves depend on notonly the principal curvature but also other factors. So we adopt a method to computethe width by considering the principal curvature of points, the dense of grid points andother factors.Sketch formed by feature belts not only contains more shape informationof3D models, but also make the sketch looks more real, feels more rich stereo.Secondly, when extract the features of sketch,[sketch-based3D shaperetrieving2012] adopt the method what is the combination of GALIF features andBOF model.GALIF feature can depict the local information of sketch well result fromit has Gabor filters in different directions. However, this method will lack the globalspatial position feature information of sketch. Therefore, in order to extract the spatiallocation information of sketch, we combine the GALIF feature extraction algorithmand Pyramid matching algorithm.Pyramid matching algorithm divide the sketch into three scales, and thefeatures are extracted in each scale,finally the extracted features in different scales areseries connected together to form a long vector, what we called the PGALIF feature.PGALIF features add the global spatial information of sketch to the final feature.when calculate the similarity of the sketch, we calculate the distance of PGALIFfeature by Pyramid matching algorithm. The distance obtained by this process candescribe the similarity of sketch better.Based on the two algorithms improved above, we propose a sketch-based3Dshape retrieval system. And we do the retrieval experiments with the shrec12modellibrary which is the standard3D model library. We compare our retrieval system tomany other existing methods based on the results of retrieval experiments. The resultof experiment shows that our method has better precision-recall curve, showing thatour method improve the outcome of the3D model retrieving based on sketch.Although we have obtained some good results, the sketch-based3D shape retrievalsystem also needs further improvement.