BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficient Image Scene Analysis and Applications - Ming-Ming Cheng\
 , University of Oxford
DTSTART:20140424T140000Z
DTEND:20140424T150000Z
UID:TALK52159@talks.cam.ac.uk
CONTACT:37004
DESCRIPTION:Images remain one of the most popular and ubiquitous media for
  capturing and documenting the world around us. Developing efficient algor
 ithms for understanding such images is of great importance for many applic
 ations in computer vision and computer graphics. In this report\, I will p
 resent three algorithms for efficient image scene understanding as well as
  their applications.\n\nAutomatic estimation of salient object regions acr
 oss images\, without any prior assumption or knowledge of the contents of 
 the corresponding scenes\, enhances many computer vision and computer grap
 hics applications. We introduce a regional contrast based salient object e
 xtraction algorithm\, which simultaneously evaluates global contrast diffe
 rences and spatial weighted coherence scores. Experimental results on famo
 us benchmarks demonstrated that our algorithm consistently outperforms exi
 sting salient object detection and segmentation methods\, yielding higher 
 precision and better recall rates. The proposed method\, which do not requ
 ire having expensive training data annotation in advance\, provides an eco
 nomical and practical tool to analysis large scale unlabeled dataset (e.g.
  internet images). \n\n\nTraining a generic objectness measure to produce 
 a small set of candidate object windows\, has been shown to speed up the c
 lassical sliding window object detection paradigm. We proposed a novel bin
 arized normed gradients (BING) feature for objectness estimation of image 
 windows. Our novel feature enables a few atomic operations (e.g. ADD\, BIT
 WISE SHIFT\, etc.) to test the objectness score of an image window. Experi
 ments on the challenging PASCAL VOC 2007 dataset show that our method effi
 ciently (300fps on a single laptop CPU\, 1000 times faster than existing m
 ethods) generates a small set of category-independent\, high quality objec
 t windows\, yielding 96.2% object detection rate (DR) with 1\,000 proposal
 s.\n\n\nHumans describe images in terms of nouns and adjectives while algo
 rithms operate on images represented as sets of pixels. Bridging this gap 
 between how we would like to access images versus their typical representa
 tion is the goal of image parsing. In this paper we propose treating nouns
  as object labels and adjectives as visual attributes. This allows us to f
 ormulate the image parsing problem as one of jointly estimating per-pixel 
 object and attribute labels from a set of training images. We propose an e
 fficient (interactive time) solution to this problem. Using the extracted 
 attribute labels as handles\, our system empowers a user to verbally refin
 e the results. This enables hands free parsing of an image into pixel-wise
  object/attribute labels that correspond to human semantics.
LOCATION:Cambridge University Engineering Department\, LR3
END:VEVENT
END:VCALENDAR