Blur detection in image sequences recorded by a wearable camera

Blur detection in image sequences recorded by a wearable camera,10.1109/NEBC.2011.5778629,Zhen Li,Zhiqiang Wei,Robert J. Sclabassi,Wenyan Jia,Mingui S

Blur detection in image sequences recorded by a wearable camera  
BibTex | RIS | RefWorks Download
A new method based on the Discrete Cosine Transform (DCT) and the Otsu method for blur detection in image sequences is proposed in this paper. In the first step, the standard deviation (STD) and the DCT coefficients are utilized to detect blurred and homogeneous areas in each image. Then, the Otsu method is used to calculate an adaptive threshold in each segment of the image sequence. Our experiments illustrate that the proposed method is robust and effective. I. INTRODUCTION Obesity problem has become one of the most serious problems in the world, and how to control diet is an important task for people to keep healthy. In our laboratory, we have developed a wearable device which contains a camera sensor collecting visual data immediately in front of the wearer during the daytime for over ten hours (1). From the recorded image sequences, it is easily to know what people choose to eat and analyze the amount of calorie intake for one day. However, there are tons of pictures obtained in one day since the camera takes a picture very second. In order to minimize the computational effort in processing these pictures, redundant pictures have to be removed before further analysis. Because of the motion of camera and the change of ambient light in different environments, some pictures are blurred and could not provide useful information. We, therefore, need to remove these blur pictures from the recorded image sequences. Many approaches have been proposed to detect blur images. Some approaches are based on spatial domain, such as edge information(2), while others are based on frequency domain, such as DCT (Discrete Cosine Transform) (3) and wavelet(4) methods. In our application, we want to change the threshold according to different environments, so we proposed a new blur detection technique combining DCT and adaptive threshold method.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.