Browsing by Author "Suh, Bongwon"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item Automatic Thumbnail Cropping and its Effectiveness(2003-04-04) Suh, Bongwon; Ling, Haibin; Bederson, Benjamin B.; Jacobs, David W.Thumbnail images provide users of image retrieval and browsing systems with a method for quickly scanning large numbers of images. Recognizing the objects in an image is important in many retrieval tasks, but thumbnails generated by shrinking the original image often render objects illegible. We study the ability of computer vision systems to detect key components of images so that intelligent cropping, prior to shrinking, can render objects more recognizable. We evaluate automatic cropping techniques 1) based on a method that detects salient portions of general images, and 2) based on automatic face detection. Our user study shows that these methods result in small thumbnails that are substantially more recognizable and easier to find in the context of visual search. Keywords: Saliency map, thumbnail, image cropping, face detection, usability study, visual search, zoomable user interfaces UMIACS-TR-2003-39 HCIL-TR-2003-13Item CounterPoint: OZONE: A Zoomable Interface for Navigating Ontology(2001-05-10) Suh, Bongwon; Bederson, Benjamin B.We present OZONE (Zoomable Ontology Navigator), for searching and browsing ontological information. OZONE visualizes query conditions and provides interactive, guided browsing for DAML (DARPA Agent Markup Language) ontologies. To visually represent objects in DAML, we define a visual model for its classes, properties and relationships between them. Properties can be expanded into classes for query refinement. The visual query can be formulated incrementally as users explore class and property structures interactively. Zoomable interface techniques are employed for effective navigation and usability. (Cross-referenced as UMIACS-TR-2001-16)Item IMAGE MANAGEMENT USING PATTERN RECOGNITION SYSTEMS(2005-04-20) Suh, Bongwon; Bederson, Benjamin B.; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)With the popular usage of personal image devices and the continued increase of computing power, casual users need to handle a large number of images on computers. Image management is challenging because in addition to searching and browsing textual metadata, we also need to address two additional challenges. First, thumbnails, which are representative forms of original images, require significant screen space to be represented meaningfully. Second, while image metadata is crucial for managing images, creating metadata for images is expensive. My research on these issues is composed of three components which address these problems. First, I explore a new way of browsing a large number of images. I redesign and implement a zoomable image browser, PhotoMesa, which is capable of showing thousands of images clustered by metadata. Combined with its simple navigation strategy, the zoomable image environment allows users to scale up the size of an image collection they can comfortably browse. Second, I examine tradeoffs of displaying thumbnails in limited screen space. While bigger thumbnails use more screen space, smaller thumbnails are hard to recognize. I introduce an automatic thumbnail cropping algorithm based on a computer vision saliency model. The cropped thumbnails keep the core informative part and remove the less informative periphery. My user study shows that users performed visual searches more than 18% faster with cropped thumbnails. Finally, I explore semi-automatic annotation techniques to help users make accurate annotations with low effort. Automatic metadata extraction is typically fast but inaccurate while manual annotation is slow but accurate. I investigate techniques to combine these two approaches. My semi-automatic annotation prototype, SAPHARI, generates image clusters which facilitate efficient bulk annotation. For automatic clustering, I present hierarchical event clustering and clothing based human recognition. Experimental results demonstrate the effectiveness of the semi-automatic annotation when applied on personal photo collections. Users were able to make annotation 49% and 6% faster with the semi-automatic annotation interface on event and face tasks, respectively.