Hand-Sketching Contour Based Image Retrieval on Mobile Device

Traditional low level features based image retrieval techniques usually have some difficulties on understanding the image similarity from the high level semantic information.To overcome this issue,under the deep-learning framework,a novel hand-sketching contour based image retrieval method on mobile devices is presented in this paper by combining image classification and exact retrieval steps.Firstly,a neural network of image classification is built including the input layer,the hidden layer and the Softmax output layer,which would be trained by image dataset.It will tell which class the input contour image belongs to after training and gets the classification label.Secondly,the VGG16 model and ResNet50 model can be loaded,by which the exact image features of each class can be extracted.Finally,a map between the combinational feature vectors and the image classification labels can be built for the purpose of image retrieval on mobile devices.Based on the C/S structure,the proposed image retrieval system would exchange data with server automatically after mobile device got the contours of input hand-sketching images.And according to the feature index and network model,the server would return the retrieval results.Using the VGG16 model and ResNet50 model loaded with Keras framework,our approach can retrieve images generated by hand-sketching contours efficiently and conveniently.