To improve the coding performance of depth maps, 3D-HEVC includes several new depth intra coding tools at the expense of increased complexity due to a flexible quadtree Coding Unit/Prediction Unit (CU/PU) partitioning structure and a huge number of intra mode candidates. Compared to natural images, depth maps contain large plain regions surrounded by sharp edges at the object boundaries. Our observation finds that the features proposed in the literature either speed up the CU/PU size decision or intra mode decision and they are also difficult to make proper predictions for CUs/PUs with the multi-directional edges in depth maps. In this work, we reveal that the CUs with multi-directional edges are highly correlated with the distribution of corner points (CPs) in the depth map. CP is proposed as a good feature that can guide to split the CUs with multi-directional edges into smaller units until only single directional edge remains. This smaller unit can then be well predicted by the conventional intra mode. Besides, a fast intra mode decision is also proposed for non-CP PUs, which prunes the conventional HEVC intra modes, skips the depth modeling mode decision, and early determines segment-wise depth coding. Furthermore, a two-step adaptive corner point selection technique is designed to make the proposed algorithm adaptive to frame content and quantization parameters, with the capability of providing the flexible tradeoff between the synthesized view quality and complexity. Simulation results show that the proposed algorithm can provide about 66% time reduction of the 3D-HEVC intra encoder without incurring noticeable performance degradation for synthesized views and it also outperforms the previous state-of-the-art algorithms in term of time reduction and $\Delta $ BDBR.
Since its introduction in 2003, the influence maximization (IM) problem has drawn significant research attention in the literature. The aim of IM, which is NP-hard, is to select a set of $k$ users known as seed users who can influence the most individuals in the social network. The state-of-the-art algorithms estimate the expected influence of nodes based on sampled diffusion paths. As the number of required samples has been recently proven to be lower bounded by a particular threshold that presets tradeoff between the accuracy and the efficiency, the result quality of these traditional solutions is hard to be further improved without sacrificing efficiency. In this article, we present an orthogonal and novel paradigm to address the IM problem by leveraging deep reinforcement learning (RL) to estimate the expected influence. In particular, we present a novel framework called deeP reInforcement leArning-based iNfluence maximizatiOn (PIANO) that incorporates network embedding and RL techniques to address this problem. In order to make it practical, we further present PIANO-E and PIANO $\text{@}\langle d\rangle $ , both of which can be applied directly to answer IM without training the model from scratch. Experimental study on real-world networks demonstrates that PIANO achieves the best performance with respect to efficiency and influence spread quality compared to state-of-the-art classical solutions. We also demonstrate that the learned parametric models generalize well across different networks. Besides, we provide a pool of pretrained PIANO models such that any IM task can be addressed by directly applying a model from the pool without training over the targeted network.
Fire has caused great losses to human beings. However, there are many problems in traditional fire detection methods.Considering the instability and high rate of erroneous recognition with these methods ,a flame recognition algorithm based on LVQ neural network is proposed in this paper. The basic characteristics and some information of the flame are analyzed.Moreover,the LVQ neural network technology is used to achieve fire detection.First, the suspicious targets of the image are extracted by flame color features.After the image morphological processing,the circular value is calculated and the interference regions with larger circular degree values is eliminated.Then,the dynamic features of the flame are extracted from the continuous frame. The area of fire will increase gradually and the image shows a continuous increase in high brightness area. The sharp corners of the flame are characterized by elongate and its number changes irregularly.Finally,the structure of the LVQ neural network, the designed of the input and output layers have been concluded. On this basis, a flame recognition algorithm based on LVQ neural network has been designed and a series of fire image experiments have been conducted.The experiment shows that the recognition accuracy of the algorithm reaches 96%.
Objective
To further reduce the registration error of ROSA by the application of corrective registration for fiducial osseous marker.
Methods
Thirty-four patients, including 33 Parkinson’s disease patients and 1 essential tremor patient, were admitted to Department of Neurosurgery, General Hospital of Shenyang Military Area Command from November 2016 to January 2017 and underwent DBS (deep brain stimulation) operation assisted by ROSA. Each of them was registered by contact with 5-point titanium alloy markers. Each patient received 4 registration plans which were respectively named group A, group B, group C and group D. The virtual probe circles in group A were tangent to the end of the marker image on CT. The virtual probe circles in group B, group C and group D were individually moved by 0.25 mm, 0.50 mm and 0.75 mm to the marker tip along the long axis based on group A. The registration errors in each group were documented and 4 sets of error data were analyzed by variance analysis.
Results
The registration errors in group A ranged from 0.47 mm to 0.77mm, with an average of 0.60±0.08 mm. The registration errors in group B ranged from 0.15 mm to 0.68 mm, with an average of 0.39±0.12 mm. The registration errors in group C ranged from 0.11 mm to 0.39 mm, with an average of 0.27±0.08 mm. The registration errors in group D ranged from 0.22 mm to 0.66 mm, with an average of 0.43±0.10 mm. The population means in the 4 groups were not completely equal by variance analysis (F=68.024, P<0.001). Significant difference was identified in registration errors between any 2 groups except for that between the group B and D. The registration error in group C was the smallest and that in group A was the largest. The registration errors in group B and D were between them.
Conclusions
Registration error decreases when the virtual probe circle is moved by 0.50 mm to the tip of markers along the long axis of markers in the DBS operation assisted by ROSA. The modified registration plan could improve the accuracy of DBS operation.
Key words:
Deep brain stimulation; Robotics; Registration error
Since its introduction in 2003, the influence maximization (IM) problem has drawn significant research attention in the literature. The aim of IM is to select a set of k users who can influence the most individuals in the social network. The problem is proven to be NP-hard. A large number of approximate algorithms have been proposed to address this problem. The state-of-the-art algorithms estimate the expected influence of nodes based on sampled diffusion paths. As the number of required samples have been recently proven to be lower bounded by a particular threshold that presets tradeoff between the accuracy and efficiency, the result quality of these traditional solutions is hard to be further improved without sacrificing efficiency. In this paper, we present an orthogonal and novel paradigm to address the IM problem by leveraging deep learning models to estimate the expected influence. Specifically, we present a novel framework called DISCO that incorporates network embedding and deep reinforcement learning techniques to address this problem. Experimental study on real-world networks demonstrates that DISCO achieves the best performance w.r.t efficiency and influence spread quality compared to state-of-the-art classical solutions. Besides, we also show that the learning model exhibits good generality.
Objective
To investigate the intraoperative threshold change of abnormal muscle response (AMR) and its relationship to clinical outcome in microvascular decompression (MVD) for hemifacial spasm (HFS).
Methods
A retrospective analysis was conducted on 67 HFS patients who underwent AMR monitoring during MVD at Neurosurgery Department, General Hospital of Shenyang Military Command, from June 2015 to March 2016. The clinical efficacy was evaluated at two time points: immediately following operation and 1 year post surgery. Threshold change of AMR from pre-decompression to decompression was documented and its relationship to clinical outcomes was analyzed.
Results
The AMR threshold in 67 cases before skin incision was 0.1-19.0 mA (median: 3.6 mA). The AMR disappeared completely in 43 cases after decompression and remained in 24 cases. The threshold of remaining AMR after decompression was 2.2-84.0 mA (median: 18.7 mA). The threshold increased less than 1 time in 9 cases and at least 1 time in 15 cases. All patients were followed up for 1 year and 60 cases (89.6%) were cured and 7 cases (10.4%) were not. Out of 43 cases with disappearance of AMR, 42 (97.7%) were cured which included 40 (93.0%) with immediate remission post surgery. Among the 24 cases whose AMR did not disappear after decompression, 18 (75.0%) were cured including 10 with immediate remission and 8 with delayed effect from 10 day to 11 months (median: 2.5 months) post surgery. Immediate and 1-year postoperative clinical outcomes were significantly different between the group with AMR disappearance and that without AMR disappearance (both P<0.05). In the group with remaining AMR, the curing rate was higher in cases with threshold increase of AMR of at least 1 time compared with that in those with threshold elevation of less than 1 time, which were 14/15 and 4/9, respectively. The difference was statistically significant (P=0.015).
Conclusions
During MVD for HFS, the disappearance of AMR seems to be associated with better outcome and delayed remission is more likely to occur in those without AMR disappearance. The threshold increase of at least 1 time post decompression compared with preoperative (prior to skin incision) level might be related to better clinical outcome.
Key words:
Hemifacial spasm; Microvascular decompression; Abnormal muscle response; Threshold; Prognosis
Image segmentation is an important issue in computer vision. The methods based on Fuzzy C-Means (FCM) algorithms have gained success. However, these approaches deal with each pixel as a separate object, which will ignore the spatial information among these pixels. This paper proposes an approach which combines the Fuzzy C-Means algorithm and Graph Cut Theory both for gray and color image segmentation. We adopt the Turbopixel algorithm to split the color image into varied small regions called superpixels for presegmentation and extract color histogram features from the superpixels. Based on color histogram feature, we use FCM to make the original clusters. Then we build a graph model, and use maximum flow algorithm to get the minimum cut, namely the initial segmentation result of the image. Finally, we use a recursive process to achieve the result of image segmentation. The key point of our approach is building a great graphical model and utilizing the existing binary segmentation model to solve the multi-value segmentation. Experimental results show that our approach can obtain good segmentation results comparing with FCM only under different parameters setup and binary segmentation.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)