Papers by Keyword: Texture Analysis

Abstract: Since it is difficult to locate license plate quickly and precisely under complicated background and different illumination conditions, therefore, this paper proposes a high-speed and precise approach to solve such kind of problem on the basis of texture analysis and projection method. The approach utilizes both rich textural feature and projective statistical feature of characters in LP region to carry out coarse and fine license plate location. The experiment shows that, under different illumination, car models and background conditions, such approach is proved to be fast, precise and practical.

Abstract: In this paper, texture analysis was used to discriminate digital chest radiographs of pneumoconiosis patients from normal ones. First, lung fields in each chest radiograph were segmented by using the morphological reconstruction and Otsu-thresholding. Second, several texture features based on the histogram and co-occurrence matrix of grey levels were extracted. Finally, a neural network based classifier was trained with features extracted from 66 chest images to distinguish pneumoconiosis patients from normal cases. Another 29 images were used to assess the diagnosis performance of the classifier, giving an overall accuracy of 79.3%.

Abstract: Conditions of tool wear can be indirectly reflected by the texture features of the machined image due to the direct contact between the work piece and tool in cutting process. In this paper, a new texture analysis method is proposed based on the Hough transform and Run length statistics. In order to remove non uniform illumination and noise, the original machined images are preprocessed by contrast enhancement algorithm and adjacent region average method, and then the edge images are obtained by a canny edge detector. Hough transform is then applied to the edge images to detect all line segments. The length characteristics of line segments are detected by using Run length statistics method. The average length and angle characteristics of edge images are used to determine the tool wear. Through our experiments, we found a high degree of correlation between texture features of machined images and tool wear. The combination of Hough transform and Run length statistics method improve the monitoring performance.

Abstract: Austenitic stainless steel with submicron gain size has been attracted due to fine structural control of mechanical properties. In order to achieve a submicron grain size, meta-stable austenitic steel AISI 304 is severely cold deformed and then annealed to different heat treatment cycles. The heat treated samples were then tested for metallurgical phase change, texture components and hardness. It was observed that at 750°C, all the martensite transformed completely into austenite. Further, at the same temperature, it was observed that the texture component {221}<232> was the dominant texture component.

Abstract: The analysis of deformed and recrystallized zirconium, used in nuclear industry is presented. The main purpose of the present work is to describe and analyze changes in texture, microstructure and misorientation profile, which are observed during a complete thermomechanical treatment. Zirconium samples were channel compressed till various degrees of deformation. The samples were then annealed. In both deformed and recrystallized states, topological maps were measured using the well known EBSD (Electron Backscatter Diffraction) technique. The obtained data were necessary for further analysis which consisted of several approaches: quantification of textures, identification of principal components, volume fractions, misorientations, grain size and IQ histograms. For the misorientation analysis, several parameters have also been tested (ie. correlated misorientation profile, grain average misorientation, kernel, etc.). Based on this complete set of data a scenario is proposed to explain the observed microstructural evolutions.

Abstract: Automatic ship detection from remote sensing images is very important as a variant of applications such as harbor management, cargo shipping, marine rescue and naval warfare will call for the aids of the analysis of these images. This paper focuses on the processing of space-born optical images (SDSOI). With the continuous development of photography technology, high-resolution remote sensing images are produced with extremely high speed, but still lack of an effective and swift method to automatically process them and get an applicable result. The whole work flow is based on three modules. First, separating land and sea with threshold segmentation, texture segmentation and region-growth and hollow-filling algorithm, and extract the sea region as ROI. Second, apply contrast box algorithm to the ROI to get the candidates of targets. Thirdly, use shape analysis to delete some simple false candidates, and use the saliency map algorithm to eliminate possible influence of clouds. Experimental results of a series of optical remote sensing images captured by satellites indicate that our approach is effective and swift in dealing with high resolution SDSOI, obtains a satisfactory ship detection miss rate and alarm rate.

Abstract: Poyang Lake is the largest freshwater lake in China which located in the Jiangxi province. Flood happened around the lake almost each summer. Remote sensing technology is an effective way to monitoring flood at large scale. This study employs images of Advanced Synthetic Aperture Radar (ASAR) and Landsat Enhance Thematic Mapping Plus (ETM+) to map the flooding area in Poyang Lake. Two ASAR scenes are adopted in this study: one was acquired in flooding period on 1st July 2010, and another was acquired on 9th September 2010 in dry season. Water body is extracted using threshold method, and flood inundation area is detected using the two images. Texture analysis of ASAR data and ETM+ indices are synchronously used in classification. Texture layers of ASAR data are computed, and Normalized Difference Vegetation Index (NDVI) and Modified Normalized Difference Water Index (MNDWI) are also calculated using ETM+ data acquired in the dry season. NDVI and MNDWI derived from ETM+ data are sensitive to soil moisture and vegetation cover., therefore, these two classes are more separable from dry and non-flooded objects than they are in a “ASAR only” classification. The indices derived from ETM+ are used as additional layers for water body classification. Supervised classification is implemented on both ASAR dataset and ASAR+TM data using Maximum Likelihood classifier. Results suggest the classification of ASAR and ETM+ combined data is more homogeneous because the indices reduced speckle-noise, and features can be clearly separated.

Abstract: The relevance of EBSD-based investigations for statements on the macroscopic or mesoscopic behavior of materials is critically relying on the statistical representativeness of the data. Particularly, the statistical reliability of the EBSD-based results (e.g. texture, phase fraction or grain size) remains an open question since the areas observed by the EBSD technique are quite small compared to XRD techniques. It has already been shown that covering larger areas and probing more grains with the help of large step sizes is beneficial in terms of representativeness [1]. On the other hand, small step sizes are beneficial in terms of grain reconstruction and data clean-up. However, step sizes significantly smaller than the average grain size of the material lead to either covered areas or number of probed grains being too small to be representative or to very large datasets and correspondingly long measurement times. In this contribution, the benefits of a new mapping technique [1] that joins the advantages of large and small step size measurements will be demonstrated. The representativeness of the EBSD datasets obtained by classical and this new mapping techniques were compared by calculating the pole figure symmetries of a TRIP steel. The results show that the proposed mapping technique significantly improves the reliability and representativeness of EBSD-based texture measurements.

Abstract: The main purpose of this work was to study the applicability of an image texture analysis method, namely, the grey level co-occurrence matrix (GLCM) method for the examination of the smoothness of the images of a turned surface. The effect of the variation of the pixel pair spacing (pps) on the construction of the GLCM was also considered and then, contrast and homogeneity were calculated from the GLCMs which served as texture descriptors for the quality of the machined surface. Finally, the variation of these texture descriptors with cutting time was analyzed and compared with the variation of tool wear and surface roughness with cutting time.

Abstract: Texture surface analysis is very important for machine vision system. We explore Gray Level Co-occurrence Matrix-based 2^nd order statistical features to understand image texture surface. We employed several features on our ground-truth dataset to understand its nature; and later employed it in a building dataset. Based on our experimental results, we can conclude that these image features can be useful for texture analysis and related fields.