{"title":"Building Better Formlet Codes for Planar Shape","authors":"A. Yakubovich, J. Elder","doi":"10.1109/CRV.2014.19","DOIUrl":null,"url":null,"abstract":"The GRID/formlet representation of planar shape has a number of nice properties [4], [10], [3], but there are also limitations: it is slow to converge for shapes with elongated parts, and it can be sensitive to parameterization as well as grossly ill-conditioned. Here we describe a number of innovations on the GRID/formlet model that address these problems: 1) By generalizing the formlet basis to include oriented deformations we achieve faster convergence for elongated parts. 2) By introducing a modest regularizing term that penalizes the total energy of each deformation we limit redundancy in formlet parameters and improve identifiability of the model. 3) By applying a recent contour remapping method [9] we eliminate problems due to drift of the model parameterization during matching pursuit. These innovations are shown to both speed convergence and to improve performance on a shape completion task.","PeriodicalId":385422,"journal":{"name":"2014 Canadian Conference on Computer and Robot Vision","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2014-05-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2014 Canadian Conference on Computer and Robot Vision","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/CRV.2014.19","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 2
Abstract
The GRID/formlet representation of planar shape has a number of nice properties [4], [10], [3], but there are also limitations: it is slow to converge for shapes with elongated parts, and it can be sensitive to parameterization as well as grossly ill-conditioned. Here we describe a number of innovations on the GRID/formlet model that address these problems: 1) By generalizing the formlet basis to include oriented deformations we achieve faster convergence for elongated parts. 2) By introducing a modest regularizing term that penalizes the total energy of each deformation we limit redundancy in formlet parameters and improve identifiability of the model. 3) By applying a recent contour remapping method [9] we eliminate problems due to drift of the model parameterization during matching pursuit. These innovations are shown to both speed convergence and to improve performance on a shape completion task.