In the existing bilateral filtering algorithm, the domain parameters and range parameters need to be predefined. Parameters of a bilateral filter are fixed and cannot guarantee to be optimal. A new adaptive bilateral filtering (ABF) is proposed in this paper. The ABF obtains the domain parameters by estimating the local object scale to minimize edge blurring. The range parameters are set adaptively according to noise variance estimated in smooth areas of a sub image. The method can improve the filtering performance. To filter out strong noise, the value of domain parameters is increased. ABF avoids setting parameters solely by experience, and the domain parameters are set adaptively according to the local image features. ABF can improve the noise filtering ability and reserves edges. Experiments show that the adaptive bilateral filter is superior to traditional bilateral filters, anisotropic diffusion filters, and modified bilateral filters in both subjective and objective evaluations.
This paper proposes a set of SVM classification methods based on fusion of gray scale and texture features. A set of experiments are carried out using the SVM classifiers with feature fusion. Both qualitative and quantitative approaches are applied to assess the classification results. Experimental results demonstrate that the proposed approach is effective for SAR image classification with accuracy higher than those obtained by using single texture feature based algorithms.
A depth map compression method oriented to virtual view rendering is proposed to reduce distortion of virtual view synthesis in a free-viewpoint television (FTV) system. The distortion is caused by quantization of depth data and inaccuracy of depth estimation algorithm. According to the statistics of depth data, a depth down/up sampling techniques based on the non-uniform B-spline curve is proposed. The depth map is down-sampled and coded with MVC. In the de-coding side, the depth map is up-sampled. To deal with the serration phenomenon in edges of the decoded depth image, a bilateral filtering is used to align the object boundaries. The experimental results evaluated by objective and subjective methods indicate that the proposed method can reduce bit-rate of depth coding and achieve better rendering quality. It can improve high-accurate reconstruction of synthesized images at the receiver of FTV systems.
This paper proposes an effective method to be used in fast detection of small surface defects. The space domain gradient method is useful to enhance the surface image. However, we show that the Otsu method cannot produce satisfied result in segmenting small defects in a large surface image. To detect small defects and improve the performance of Otsu method, an algorithm based on the distribution of variances of image blocks is developed to search of small defect regions. Analysis and experiments show that the proposed method can be applied for fast detection of small surface defects.
Based on the characteristics of high spatial resolution images, an object-oriented approach to the building extraction is proposed. The method takes the object as a processing unit to perform classification using the rich information such as the spectrum, the object shape, and the implicit space semantic relations embedded in the images. The classification result is optimized based on the characteristics of the spatial relationship. Edges between different categories are modified using mean filtering. Experiments show that the object-oriented method can provide relatively complete extraction of the building, and improve precision of the extraction.
By computing eigenvalues of special square matrices derived from the incidence matrices of Petri net models, some important structural properties of Petri nets are analyzed. According to the differences between two types of nets, i.e., choice-free and link-free nets, and non-choice-free and non-link-free nets, two methods are used to transform them into the same class of square matrices. We then obtain sufficient conditions for structure boundedness, conservativeness, repetitiveness and consistency based on the theory of M-matrices. An example of radar model is given to show application in analyzing Petri net structures.
Mobile station performs a handoff when it moves out of the range of one access point and tries to connect to the next. The delay and lost packets caused by the handoff process affect the network quality of service. Taking subway wireless communication system as an application scenario, the number of channels used is reduced, which decreases the scan delay effectively. An algorithm called Layer2-Buff is proposed to solve the problem of lost packets, which sets up buffers for STA, and used to save data that cannot be sent successfully. Test results show that Layer2-Buff reduces the packet loss rate to zero in an automatic train control system, and extra delay to the network is tolerable. This algorithm has been applied to a pre-research project of subway signal systems.
A distributed algorithm is proposed for resource allocation in self-organized home eNodeB (HeNB)networks. In this algorithm, each HeNB selects component carriers (CC) based on the network status. If a reselection request is received from neighboring cells, the HeNB will reselect the CC in the backup CC list, and release the CC requested by neighbors to enhance possibility of successful carrier selection. Each HeNB then adaptively performs power optimization in active CCs to reduce inter-cell interference and enhance the system capacity. Simulation results show that this resource and power allocation algorithm can enhance average user throughput and reduce the outage rate.
