Streetscape picture retrieval method based on deep global features

• Main Authors: HUANG LIHENG, FANG TAO, CHU TIANYOU, LI HUIFANG, LUO FENGLAN, TAN HUANGYUAN, CHEN YUMIN
• Format: Patent
• Language: Chinese; English
• Published: 15.09.2020
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses a streetscape picture retrieval method based on depth global features. According to the streetscape picture retrieval method, streetscape picture features are encoded by usinga deep convolutional neural network method; streetscape pictures are expressed through a single long feature vector; primary picture sorting is realized through similarity comparison of Euclidean distances; and finally, a primary sorting result is resorted through SIFT features. According to the streetscape picture retrieval method, the problem that traditional features cannot effectively expressimages is solved; and through training of a specific street view landmark data set, the sensitivity of the network to building features is enhanced, and the adaptability of the network to street viewimage retrieval is enhanced, and effective feature abstraction and expression can be more easily performed on street view images, and the retrieval speed can be increased through the image expressionmode of a single feature vec