Image Retrieval Algorithms For Design Patent Based On Multi-Feature Fusion

Product design plays a more and more important role in the market competition. Product design patents retrieval technology not only helps designers find inspiration, but also can help companies avoid the risk of infringement. It has great significant meaning for the protection of intellectual property.This thesis puts forward two retrieval algorithms for design patent images based on multi-feature fusion according to the characteristics of design patent images. The multi-feature fusion algorithm based on Gaussian descriptor extracts shape features by using Hu invariant moments and Gaussian descriptor, and texture features by using Gabor wavelet. The algorithm obtains the retrieval results according to the similarity distance of multi-feature fusion in the retrieval process. The multi-feature fusion algorithm based on edge direction histogram extracts shape features by using structure feature and edge direction histogram, and texture features by using Garbor wavelet and statistical characteristics of image blocks. Several geometric features are used to roughly classify the images in the image library. Then image features are transformed symmetrically. And the similarity distance between pretransformed features and transformed features are calculated. The actual similarity distance is the shorter one. Finally retrieval results are achieved according to similarity distance.This thesis analyses the performance of several multi-feature fusion algorithms for image retrieval. The experimental results indicate that the multi-feature fusion based on edge direction histogram is most feasible for design patent image retrieval. Finally this thesis realizes a simulation image retrieval system for design patents based on this algorithm.