| Summary: | 碩士 === 元智大學 === 資訊工程學系 === 100 === Logos are specially designed marks that identify goods, services, and organizations using notable characters, graphs, signals, and colors. Identifying logos can facilitate scene understanding, intelligent navigation, and object recognition. Although many logo recognition methods have been proposed for printed logos, few are specifically designed for logos in photos. Moreover, most existing recognition methods for logos in photos adopt codebook-based approaches. A codebook-based method is concerned with generation of visual words for all the logo models, which requires codebook reconstruction when new logos are added, yielding reconstruction overhead. On the other hand, logos in natural scenes have perspective tilt and non-rigid deformation leading to main challenging. Therefore, the objective of this study is to develop extendable yet discriminative model-based logo detection.
The proposed logo detection is based on SVM (Support Vector Machine) using HOGE (Edge-based Histograms of Oriented Gradient Histogram) as features through multi-scale sliding window scanning. The anti-distortion ASIFT (Affine Scale Invariant Feature Transform) are then used for logo verification with constraints on ASIFT matching pairs and neighbors. Experimental results using Flickr-Logo database confirm that the proposed method has higher retrieval and precision accuracy compared to existing model-based methods.
|
|---|
