Grass, Scrub, Tree, Random Forest and Human in the Loop-Automatic habitat classification based on digital photographs

The classification of habitats is important for developing our understanding of the natural world and for monitoring the environment and biodiversity. Currently, habitat classification is performed by human surveyors ─ a laborious, expensive and subjective process. In this talk, I will present our work in developing automatic habitat classification solutions based on ground photographs. We employ image analysis methodologies from the multimedia and computer vision literature and formulate habitat classification as a fine-grained visual categorization problem. We treat digital photograph based habitat classification as an image-labeling problem and have developed an image-annotation framework that uses a random forest to automatically label a digital photo with habitat classes. As automatic visual recognition is still very hard whilst humans can perform visual recognition with the greatest ease, we have developed a human in the loop approach to enhancing the accuracy of automatic habitat classification. Using a crowd-sourcing mechanism, we extract medium-level knowledge about the photographs by asking humans a set of 17 questions about the visual appearances of the image that can be easily answered by non-ecologists. This medium-level knowledge is then fused with the low-level visual features in a random forest to improve the accuracy of habitat classification. Digital photographs taken with modern cameras (smartphones included) contains location information (GPS) and based on the reasonable assumption that neighboring areas should have similar ecological properties, we use the GPS information of the photos together with the random forest outputs to further improve the habitat classification performances. We have constructed a geo-referenced habitat photograph database containing over 1000 high-resolution ground photographs that have been manually annotated by habitat classification experts. This database has been specifically designed for the development of multimedia analysis techniques for ecological applications and is publicly available. I will show experimental results to illustrate that with these multimedia analysis techniques, it is possible to annotate with a reasonable degree of confidence four of the main habitat classes: Woodland and Scrub, Grassland and Marsh, Heathland and Miscellaneous. This is a joint work with Dr Mercedes Torres.

by Prof. Guoping Qiu, School of Computer Science, University of Nottingham, Nottingham, UK.

Guoping Qiu is Professor of Visual Information Processing in the School of Computer Science at the University of Nottingham, Nottingham, UK. He joined Nottingham in October 2000 as a Lecturer (assistant professor) and was subsequently promoted to a Reader (associate professor) and a Chair. His research interests include image processing, pattern recognition, multimedia processing, machine learning and their applications to real world problems including habitat classification, digital pathology, content-based image retrieval and automatic image tagging. He has particular expertise in high dynamic range (HDR) imaging and has developed several highly successfully techniques that are now routinely used in many digital photography software and smartphone camera apps. He has published over 170 papers in these areas and holds several European and US patents. He has won several prizes including a best paper award at the 18th International Conference on Pattern Recognition (ICPR2006). He has taught in universities in the UK and Hong Kong and consulted for multinational companies in Europe, Hong Kong and China. Recently, he took up a secondment to the University of Nottingham’s China campus where he is leading the School of Computer Science and the International Doctoral Training Centre (IDIC), a £17 million investment venture to train PhD students in the areas of energy technologies and digital economy.  His current research projects include building a high dynamic range video camera and a 3D digital microscope. He and his students are working on automatic traffic video analysis, video based crowd detection and analysis, near duplicate image and video discovery and their applications to smart city development amongst others.