Query-by-example HDR image retrieval based on CNN

Due to the expension of High Dynamic Range (HDR) imaging applications into various aspects daily life, an efficient retrieval system, tailored this type data, has become a pressing challenge. In paper, reliability Convolutional Neural Networks (CNN) descriptor and its investigation for HDR image are studied. The main idea consists in exploring use CNN determine descriptor. Specifically, Perceptually Uniform (PU) encoding is initially applied content map luminance values perceptually uniform scale. Afterward, features, using Fully Connected (FC) layer activation, extracted classified by applying Support Vector Machines (SVM) algorithm. Experimental evaluation demonstrates that descriptor, VGG19 network, achieves satisfactory results describing images on public available datasets such as PascalVoc2007, Cifar-10 Wang. experimental also show after PU processing, more descriptive than those directly from contents. Finally, we superior performance proposed method against recent state-of-the-art technique.