List of Articles Manifold Learning

• Open Access Article
  • Abstract Page
  • Full-Text
  
  1 - A new Sparse Coding Approach for Human Face and Action Recognition
  Mohsen Nikpoor Mohammad Reza Karami-Mollaei Reza Ghaderi
  10.7508/jist.2017.17.003
  Sparse coding is an unsupervised method which learns a set of over-complete bases to represent data such as image, video and etc. In the cases where we have some similar images from the different classes, using the sparse coding method the images may be classified into More

  Sparse coding is an unsupervised method which learns a set of over-complete bases to represent data such as image, video and etc. In the cases where we have some similar images from the different classes, using the sparse coding method the images may be classified into the same class and devalue classification performance. In this paper, we propose an Affine Graph Regularized Sparse Coding approach for resolving this problem. We apply the sparse coding and graph regularized sparse coding approaches by adding the affinity constraint to the objective function to improve the recognition rate. Several experiments has been done on well-known face datasets such as ORL and YALE. The first experiment has been done on ORL dataset for face recognition and the second one has been done on YALE dataset for face expression detection. Both experiments have been compared with the basic approaches for evaluating the proposed method. The simulation results show that the proposed method can significantly outperform previous methods in face classification. In addition, the proposed method is applied to KTH action dataset and the results show that the proposed sparse coding approach could be applied for action recognition applications too. Manuscript profile
• Open Access Article
  • Abstract Page
  • Full-Text
  
  2 - Improving Pose Manifold and Virtual Images Using Bidirectional Neural Networks in Face Recognition Using Single Image per Person
  F. Abdolali S. A. Seyed Salehi
  In this article, for the purpose of improving neural network models applied in face recognition using single image per person, a bidirectional neural network inspired of neocortex functional model is presented. In the proposed model, recognition is not performed in a si More

  In this article, for the purpose of improving neural network models applied in face recognition using single image per person, a bidirectional neural network inspired of neocortex functional model is presented. In the proposed model, recognition is not performed in a single stage, but via two bottom-up and top-down phases and the recognition results of first stage is used for model adaptation. We have applied this novel adapting model in combination with clustering person and pose information technique to separate person and pose information and to estimate corresponding manifolds. To increase the number of training samples in the classifier neural network, virtual views of frontal images in the test dataset are synthesized using estimated manifolds. Training classifier network via virtual images obtained from bidirectional network, gives an accuracy rate of 85.45% on the test dataset which shows 1.82% improvement in accuracy of face recognition compared to training classifier with virtual images obtained from clustering person and pose information network. Manuscript profile
• Open Access Article
  • Abstract Page
  • Full-Text
  
  3 - Left Ventricular Segmentation in Echocardiography Images by Manifold Learning and Dynamic Directed Vector Field Convolution
  N.  Mashhadi H. Behnam Ahmad Shalbaf Z. Alizadeh Sani
  Cardiac diseases are the major causes of death throughout the world. The study of left ventricular (LV) function is very important in the diagnosis of heart diseases. Automatic tracking of the boundaries of the LV wall during a cardiac cycle is used for quantification o More

  Cardiac diseases are the major causes of death throughout the world. The study of left ventricular (LV) function is very important in the diagnosis of heart diseases. Automatic tracking of the boundaries of the LV wall during a cardiac cycle is used for quantification of LV myocardial function in order to diagnose various heart diseases including ischemic disease. In this paper, a new automatic method for segmentation of the LV in echocardiography images of one cardiac cycle by combination of manifold learning and active contour based dynamic directed vector field convolution (DDVFC) is proposed. In this method, first echocardiography images of one cardiac cycle have been embedded in a two dimensional (2-D) space using one of the most popular manifold learning algorithms named Locally Linear Embeddings. In this new space, relationship between these images is well represented. Then, segmentation of the LV wall during a cardiac cycle is done using active contour based DDVFC. In this method, final contour of each segmented frame is used as the initial contour of the next frame. In addition, in order to increase the accuracy of the LV segmentation and also prevent the boundary distortion, maximum range of the active contour motion is limited by Euclidean distances between consequent frames in resultant 2-D manifold. To quantitatively evaluate the proposed method, echoacardiography images of 5 healthy volunteers and 4 patients are used. The results obtained by our method are quantitatively compared to those obtained manually by the highly experienced echocardiographer (gold standard) which depicts the high accuracy of the presented method. Manuscript profile