Content-based Image Retrieval: out of the Clutter in a Big Supermarket

Experienced decades of rapid development in digital vision sensors and storage devices, today's world is enduring a flood of digital images. Practical image retrieval system is of more value now than ever. Traditional image retrieval systems were based on the text labels manually attached to each image, which requires intensive human labors and can hardly keep up with the growing number of images. Hence retrieval systems directly based on the visual content of images make their way to the foreground. Unlike other existing content-based image retrieval systems, the image to be retrieved in this project contains extremely strong background clutters and significant scale variance. The existence of multiple copies of identical objects can also confuse the retrieval system if it relies on previous methods. With the methods proposed in this book, we build up a system which can retrieve queried objects from thousands of candidates accurately and fast. This work demonstrates and promotes the practicability of computer vision technology in object recognition applications. It can be a reference to both industrials and research.