This paper introduces the technique of graph cut into the extraction of object region and applies the corresponding result
of object region extraction into the image retrieval based on object region. The main idea of image retrieval based on
object region is to use the feature of object region instead of the feature of global image to participate in the image
retrieval. In the field of graphics there is a technique called graph cut, which can be used to figure out the contour of
object under the interaction of users. The graph cut algorithm can be used to verify the correctness of object region
extraction, and the users' input about seeds can be simulated according to the initial object region extracted. The usage of
graph cut can make the object region extracted more precisely and thus the performance of image retrieval based on
object region can be improved. Experiments show that the object region extraction algorithm based on graph cut is valid
and the subsequent image retrieval results accord with the human visual perception much more than the ones without the
usage of graph cut.
The driver assistant system has attracted much attention as an essential component of intelligent transportation systems.
One task of driver assistant system is to prevent the drivers from fatigue. For the fatigue detection it is natural that the
information about eyes should be utilized. The driver fatigue can be divided into two types, one is the sleep with eyes
close and another is the sleep with eyes open. Considering that the fatigue detection is related with the prior knowledge
and probabilistic statistics, the dynamic Bayesian network is used as the analysis tool to perform the reasoning of fatigue.
Two kinds of experiments are performed to verify the system effectiveness, one is based on the video got from the
laboratory and another is based on the video got from the real driving situation. Ten persons participate in the test and
the experimental result is that, in the laboratory all the fatigue events can be detected, and in the practical vehicle the
detection ratio is about 85%. Experiments show that in most of situations the proposed system works and the
corresponding performance is satisfying.
Integral image as an intermediate image representation can be used to calculate the sum of gray level in rectangle quickly, based on which this paper presents a novel partial matching error function based on sub block mean. The optimal sub block division is a key to the matching error function based on sub block mean and is determined to be 4 under the hierarchical block matching. Experiments show that the matching error function based on sub block mean is superior to both the full matching error function and the matching error function based on sub sampling in terms of the motion estimation quality and speed. The matching error function based on sub block mean with the content sub block division guarantees the almost same wasting time to the different matching images, which is very suitable for the real-time applications such as video compression.
This paper presents an event based soccer video retrieval method, where the scoring even is detected based on Bayesian network from six kinds of cue information including gate, face, audio, texture, caption and text. The topology within the Bayesian network is predefined by hand according to the domain knowledge and the probability distributions are learned in the case of the known structure and full observability. The resulting event probability from the Bayesian network is used as the feature vector to perform the video retrieval. Experiments show that the true and false detection rations for the scoring event are about 90% and 16.67% respectively, and that the video retrieval result based on event is superior to that based on low-level features in the human visual perception.
This paper combines the template-based method and color-based scheme to construct an adaptive skin-color model for human face detection in news videos. A heuristic rule-based decision tree is then employed to verify the resulting skin-color regions. The skin-color model comes from the sample pixels from the target video shots, so well tuned to adapt to various videos. It is a general scheme for color-segmentation, not depend on any pre-defined skin-color range. Our experiments shows that the face detection performance has been improved greatly, compared with the pure template based face detector. The face retrieval module is based on the Self-Eigenface method, where the Self-Eigenface space is constructed from the pseudo frontal faces obtained by region tracking.
In this paper one color image retrieval algorithm based on object regions is presented. First obtain each component image in the HSV space and then get the binary edge image of each component image. According to the connectedness of the edge images the object regions of the color image are extracted. During the image retrieval the sub image features corresponding to the object regions are used in the image similarity matching process in place of the global image features. Experiments show that the colorimage retrieval algorithm based on object regions is superior to that based on the global image.
Coarseness is the most fundamental textural feature and has been much investigated since early studies. This paper improves the previous coarseness algorithm on the selection of neighborhood sizes and the calculation of neighborhood average differences, and the improved coarseness algorithm is presented. Experiments show that the improved coarseness has higher texture discriminability and better rotation invariance, and that the image retrieval result based on the improved coarseness is superior to that based on the previous coarseness.
The local features are as important as global features for content-based trademark image retrieval. This paper gives a trademark image retrieval method based on the features of sub-images together with global image information. We extract the sub-images for each candidate, and take the image for the sub-image of itself, and then use the features of sub-images for retrieval. We have tested our method on an image database containing 3000 binary trademark images and use PVR-component as the evaluation measure, experiments show that using local information together with the global information, the retrieval performance of our method is better than that of retrieval method based only on global features, and the retrieval result can fit the people's visual feelings well.