Search

Image recognition based on machine learning has been widely utilised in the computer vision field. In the image recognition process, quite a few positive and negative instances are needed for effective machine learning. However, some invalid instances selected from the instance candidates, particularly for negative instances, will result in reduced image recognition accuracy and wasted resources. When making the instance selections for machine learning, if a large number of negative instances are selected that exhibit a semantic distance too close to the positive instances, the recognition accuracy will be considerably reduced. The selection of valid negative instances from the instance candidates has become a new challenge in the field of computer vision. In this paper a new method containing several different algorithms is presented, that more effectively selects valid negative instances. In this innovative process, a Wordnet subtree, semantic distance based on feature descriptors and an improved K-nearest neighbour model were all utilised. Experiments were implemented using a large database to determine the effectiveness of this method. The results have demonstrated that this method can significantly improve the recognition results.

In this paper, we provide a unified expression to obtain the conditions on the restricted isometry constant δ2s(Φ). These conditions cover the important results proposed by Candes et al. and each of them is a sufficient condition for sparse signal recovery. In the noiseless case, when δ2s(Φ) satisfies any one of these conditions, the s-sparse signal can be exactly recovered via (l1) constrained minimization.

Guiding the growth of a neurite by directing ~800 nm laser light to the leading edge of the neurite's growing region can be accomplished by controlling the position and direction in three dimensional space of a tapered optical fiber through which the light is projected. We control the position, angle and power of the laser beam to direct the growth of actin accumulations in neurites which affects their mobility.

In non-photorealistic rendering (NPR), the Chinese ink painting style rendering is a traditional NPR skill of China. In this paper, we propose a method for image-based and ink-diffusion-based Chinese ink painting NPR. Users without painting experience can also convert a normal image to a Chinese ink painting automatically. As we known, ink is an important pigment for Chinese ink painting and the various ink shade effects on the painting are produced by ink mixed with water. In addition, the ink diffusion along the boundaries is a very important aspect of Chinese ink painting. In order to realize the effects described above, we present a Chinese ink painting NPR method based on ink diffusion. We use Mean Shift based image segmentation algorithm to preprocess the input image to get regions with different tones. Then, we detect the segmentation regions' edges letting the edge points to be the start points for diffusion. We set each point an ink value which is corresponding to its gray value. At the same time, a new algorithm simulating ink diffusion is proposed to make the segmentation image look like a black-ink painting. Results in this paper demonstrate our method is promising.

Image inpainting refers to the process of reconstructing the damaged image or removal of unwanted objects in the image. We need to take the most appropriate way of inpainting the image to its original state, while ensuring that the result achieves the best artistic effect. There are some methods commonly used, such as method based on PDE or based on texture synthesis. But in cases that the damaged region is too large or a semantic fragment is completely missed, those old methods cannot do a good job. Rather than inpainting within the original image data, a way of inpainting using image content from other pictures will do help. As long as the improvement of research in image retrieval, we hope to take advantage of the good results by image retrieval with particular criteria. With the image patches from the image retrieval results, we can do a better inpainting and achieve the result more in line with the visual effects. This paper shows the algorithm and experiment results show the high efficiency and quality of the algorithm. And the unexpected result due to the random computation shows its creative accomplishment.

Image inpainting technique has been widely used for completing missing parts caused by the removal of foreground or background elements from an image in a plausible way. A novel and efficient image inpainting method is proposed in this paper. First, structure propagation is used to synthesize regions in the unknown region along these salient structures specified by user. After structure completion, a finer algorithm is used to fill in the remaining unknown regions. It can prevent from erroneous matching blocks and reduce the breaking of salient structures which human eyes are sensitive to. Experiment results show the high efficiency and quality of the algorithm.

Human motion capture and editing technologies are widely used in computer animation production. We can acquire original motion data by human motion capture system, and then process it by motion editing system. However, noise embed in original motion data maybe introduced by extracting the target, three-dimensional reconstruction process, optimizing algorithm and devices itself in human motion capture system. The motion data must be modified before used to make videos, otherwise the animation figures will be jerky and their behavior is unnatural. Therefore, motion smoothing is essential. In this paper, we compare and summarize three methods of smoothing original motion capture data.

Rhodium-attapulgite (Rh-Atta) catalyst was prepared by immobilizing Rh3on Atta via an ionic liquid (IL), 1,1,3,3-tetramethylguanidine lactate (TMGL-), followed by reduction with hydrogen at 300 °C. The resulting composite was characterized by different techniques. X-ray photoelectron spectroscopy analysis showed that the loaded rhodium on Atta existed mainly in the form of Rh0with a small amount of its oxides and was distributed uniformly on Atta with a particle size of less than 5 nm, as confirmed by transmission electron microscopy examination. X-ray diffraction analysis indicated that the layered channel structure of Atta was destroyed to some extent due to the impregnation of the IL and Rh. IR spectra indicated that the Lewis and Bronsted acid sites in Atta absorbed the TMGions chemically. The specific surface areas determined by the BET method from N2sorption isotherms decreased with the entrance of the IL and loaded Rh. The activity of the composite for cyclohexene hydrogenation was investigated, which exhibited much higher efficiency compared to those of other catalysts, and the turnover frequencies reached 2700 (mol of cyclohexene/mol of Rh)/h.

A novel nickel complex, bis[2‐methyl‐2,4‐bis(2′‐pyridyl)‐1H‐1,5‐benzodiazepine]nickel dichloride, displayed good activity for oligomerization and polymerization of ethylene with the assistance of a co‐catalyst, methylaluminoxane (MAO). The oligomers were mainly olefins from C4 to C10, while the highly branched polyethylene (PE) had molecular weights (Mw) in the range 19 000–34 000. Bimodal distributions of the PE were clearly observed by both DSC and GPC measurements, while the spherulitic structure of the PE was shown in the morphology. Copyright © 2004 Society of Chemical Industry

Summary: The effect of a surface-vacuum-evaporated ultrathin carbon layer on the bulk crystallization behavior of isotactic polypropylene (iPP) films was investigated by means of polarized optical microscopy, differential scanning calorimetry (DSC), and wide-angle X-ray diffraction. Results obtained from both isothermal crystallization and constant cooling experiments indicate that surface carbon deposition on an iPP film can remarkably increase the crystallization ability of it during the melt recrystallization process. Optical microscopy observations show that the melt recrystallization of iPP with a carbon coating starts about 10 °C earlier than the pure iPP during a re-cooling process of 1 °C · min−1. Under isothermal crystallization conditions, a much shorter induction period for nucleation is required for the carbon-coated iPP than the non-carbon-coated one. The wide-angle X-ray diffraction and DSC analyses further demonstrate a crystallinity increment of the carbon‐coated iPP with respect to its non‐carbon‐coated counterpart. This clearly indicates that the strong influence of the interfacial interaction between the vacuum‐evaporated carbon and the iPP film on its bulk crystallization can be extended deep into the bulk material.An OM micrograph of a hot-pressed partially carbon-coated (top half) iPP film.

A novel and simple method to replicate biological organizations cotton and pollen grains with high precision was proposed, in which the precursor dissolved in supercritical CO2reacted with the surface active groups and adsorbed surface water on biological templates, followed by in situSCF extraction of the byproducts and unreacted precursor, resulting in inorganic replicas faithfully copying both the macro- and microstructures of the biotemplates.

Action recognition has already been a heated research topic recently, which attempts to classify different human actions in videos. The current main-stream methods generally utilize ImageNet-pretrained model as features extractor, however it's not the optimal choice to pretrain a model for classifying videos on a huge still image dataset. What's more, very few works notice that 3D convolution neural network(3D CNN) is better for low-level spatial-temporal features extraction while recurrent neural network(RNN) is better for modelling high-level temporal feature sequences. Consequently, a novel model is proposed in our work to address the two problems mentioned above. First, we pretrain 3D CNN model on huge video action recognition dataset Kinetics to improve generality of the model. And then long short term memory(LSTM) is introduced to model the high-level temporal features produced by the Kinetics-pretrained 3D CNN model. Our experiments results show that the Kinetics-pretrained model can generally outperform ImageNet-pretrained model. And our proposed network finally achieve leading performance on UCF-101 dataset.

In this paper, we present an improved approach to recognize human action based on the BOW model and the pLSA model. We propose an improved feature with optical flow to build our bag of words. This feature is able to reduce the high dimension of the pure optical flow template and also has abundant motion information. Then, we use the topic model of pLSA (probabilistic Latent Semantic Analysis) to classify human actions in a special way. We find that the existing methods lead to some mismatching of words due to the k-means clustering approach. To reduce the probability of mismatching, we add the spatial information to each word and improve the training and testing approach. Our approach of recognition is tested on two datasets, the KTH datasets and WEIZMANN datasets. The result shows its good performance.

Image matting plays an important role in image editing and computer vision. Alpha matting is to solve the problem of softly extracting the foreground object from an image. The problem is inherently ill-posed. Existing matting approaches usually use a given trimap to estimate the alpha value for each unknown pixel. So the results rely on not only the different algorithms, but also the given trimap. In this paper, we present an easy interactive matting method based on the sample search matting. Users only require drawing some strokes or points on the foreground and background to extract the foreground object. This method can easily collect the samples available to compute the alpha value for the whole image pixels.