UNIGE document Scientific Article
previous document  unige:88091  next document
add to browser collection

SimMap: Similarity maps for scale invariant local shape descriptors

Wang, Changhu
Published in Neurocomputing. 2016, vol. 175, p. 888-898
Abstract Traditional approaches to estimate a scale invariant spatial scope for local image descriptors, a.k.a. characteristic scale, work well for intensity images. However, they fail when it comes to deal with binary images. We address this problem and propose a new method to estimate the characteristic scale of local shape descriptors. The proposed method extends the use of the distance map transform to produce similarity maps that approximate local intensity changes in binary images. We first validated our method evaluating the consistency of characteristic scales estimated across scaled instances of images; and then by comparing its performance, with respect to traditional methods, in the tasks of Content-Based Image Retrieval and shape detection in different datasets of binary images (shapes of Maya and Chinese hieroglyphs, and generic shapes). As shown by our results, the proposed similarity map produces characteristic scales that are more robust to scale variations, and leads to competitive results.
Keywords Binary imagesLocal shape descriptorsShape contextCharacteristic scaleSimilarity mapShape detection
Full text
Research groups Computer Vision and Multimedia Laboratory
Viper group
(ISO format)
ROMAN RANGEL, Edgar Francisco, WANG, Changhu, MARCHAND-MAILLET, Stéphane. SimMap: Similarity maps for scale invariant local shape descriptors. In: Neurocomputing, 2016, vol. 175, p. 888-898. doi: 10.1016/j.neucom.2015.06.093 https://archive-ouverte.unige.ch/unige:88091

408 hits

0 download


Deposited on : 2016-10-10

Export document
Format :
Citation style :