In this work we present a deep convolutional neural network using 3D convolutions for Gait Recognition in multiple views capturing spatio-temporal features. A special input format, consisting of the gray-scale image and optical flow enhance color invariance. The approach is evaluated on three different datasets, including variances clothing, walking speeds view angle. contrast to most state-of-the-art systems used able generalize gait features across large angle changes. results show...

In this work, we present a novel background subtraction system that uses deep Convolutional Neural Network (CNN) to perform the segmentation. With approach, feature engineering and parameter tuning become unnecessary since network parameters can be learned from data by training single CNN handle various video scenes. Additionally, propose new approach estimate model video. For of CNN, employed randomly 5 percent frames their ground truth segmentations taken Change Detection challenge...

The problem of human activity recognition can be approached using spatio-temporal variations in successive video frames. In this paper, a new technique is proposed multi-view videos. Initially, naive background subtraction frame differencing between adjacent frames performed. Then, the motion information each pixel recorded binary indicating existence/non-existence frame. A wise sum over all difference images view gives frequency throughout clip. classification performances are evaluated...

Active learning has gained a high amount of attention due to its ability label vast unlabeled collected earth observation (EO) data. In this paper, we propose novel active algorithm which is mainly based on employing low-rank classifier as the training model and introducing visualization support data point selection, namely, first certain wrong labeled (FCWL). The composed logistic regression loss function trace-norm parameters regularizer. FCWL selects those points whose labels are...

Dramatic growth in the volume of data made a compact and informative representation highly demanded computer vision, information retrieval, pattern recognition. Non-negative Matrix Factorization (NMF) is used widely to provide parts-based representations by factorizing matrix into non-negative factors. Since non-negativity constraint not sufficient achieve robust results, variants NMF have been introduced exploit geometry space. While these considered local invariance based on manifold...

We are dealing with large-scale high-dimensional image data sets requiring new approaches for mining where visualization plays the main role. Dimension reduction (DR) techniques widely used to visualize data. However, information loss due reducing number of dimensions is drawback DRs. In this paper, we introduce a novel metric assess quality DRs in terms preserving structure model dimensionality process as communication channel transferring points from space (input) lower one (output)....

Images recorded in turbid waters suffer from various forms of signal degradation due to light absorption, scattering and backscatter. Much the earlier work enhance color, contrast sharpness follow single-image dehazing approach atmospheric imaging literature. Requiring knowledge both range scene objects ambient lighting, techniques differ how they estimate information image regions. Moreover, some assumptions are made that hold for most images air clear waters, but often violated...

Earth observation (EO) images clustering is a challenging problem in data mining, where each image represented by high-dimensional feature vector. However, the vectors might not be appropriate to express semantic content of images, which eventually lead poor results and classification. To tackle this problem, we propose an interactive approach generate compact informative features from content. end, utilize 3-D application support user-images interactions. These interactions are used context...

Multiple object tracking is a challenging problem in computer vision due to difficulty dealing with motion prediction, occlusion handling, and re-identification. Many recent algorithms use appearance cues overcome these challenges. But using increases the computation cost notably therefore speed of algorithm decreases significantly which makes them inappropriate for online applications. In contrast, there are that only increase speed, especially cannot handle occlusions re-identify lost...

In this paper, we propose a method to cluster multiple intersected manifolds. The algorithm chooses several landmark nodes randomly and then checks whether there is an angle-constrained path between each node every other in the neighborhood graph. When points lie on different manifolds with intersection they should not be connected using smooth path, thus angle constraint used prevent connecting from one another one. resulting implemented as simple variation of Dijkstra's Isomap. However,...

Image inpainting is the task of removing undesired objects or flaws in images. This work advances an exemplar-based global optimization image algorithm. For that purpose, area iteratively refined through minimization a cost function. The outcome depends on initial values area. We compare three initialization methods with new sparsity-driven approach. Lastly, we propose wavelet contrast costs which increase quality. Wavelet contrasts reduce computational complexity comparison to histograms...

We propose a novel approach to address the problem of jointly tracking and gait recognition multiple people in video sequence. The most state art algorithms for consider cases where there is only one person without any occlusion very constrained environment. However, real scenarios such as airports, train stations, etc, are many environment that make these inapplicable. Although first each then could be solution, we argue multi-people two sub-problems can help other. Hence, joint framework...

To overcome challenges in multiple object tracking task, recent algorithms use interaction cues alongside motion and appearance features. These graph neural networks or transformers to extract features that lead high computation costs. In this paper, a novel cue based on geometric is presented aiming detect occlusion re-identify lost targets with low computational cost. Moreover, most algorithms, camera considered negligible, which strong assumption not always true leads ID Switch...

Representing images with their descriptive features is the fundamental problem in CBIR. Feature coding as a key-step feature description has attracted attentions recent years. Among proposed strategies, Bag-of-Words (BoW) most widely used model. Recently saliency been mentioned characteristic of BoW. Base on this idea, Salient Coding (SaC) introduced. Empirical studies show that SaC not able to represent global structure data small number codewords. In paper, we remedy limitation by...