Earticle

Face recognition is the process of identifying one or more people in images or videos. It is an important part of biometric, security & surveillance system, and image indexing systems. Various face recognition techniques have been proposed in literature such as: Eigen-faces, Feature based, Hidden Markov model and Neural network based techniques. The first three techniques mostly include a phase of feature extraction or preprocessing closely related to the type of image to recognize. On the other hand Neural network technique does not need specific data about the type of image, thus can be applied to any type of image and at the same time provides better accuracy. In this paper we made an effort to combine neural network technique with fuzzy logic. Our experimental result shows that combining the two provide better accuracy in comparison to other techniques mentioned above.

This text discusses the feature of the existing methods relating to content-aware image resizing in detail, compares the performance and effect between different methods, and analyzes the merits and demerits about different algorithms. The basic process in content-aware image resizing are segmented into two steps: one of the steps is the identification of image content, which is generally computed by the product of image gradient and vision significance, and it reflects the different sensitivity that human eyes may feel towards different areas in image; Present articles regarding to content-aware image resizing mainly focus on the resizing technology that is the second step. According to the traits of different methods, this article divides the methods into three sorts: based on seam carving, image deformation and multi-operation.

In this paper, the methods of correcting skew vehicle license plate and segmenting characters in plate are discussed first. An approach making use of self-organizing map (SOM) is introduced to find the tilt angle of plate which simultaneously educes seven important points with coordinates being elements of weight matrix. After necessary processing to corrected plate, a character segmentation algorithm based on the shortest distance classification is presented, which takes advantage of exactly seven points gained from SOM as class centers. In the next place, a hybrid algorithm cascading two steps of template matching is utilized to recognize Chinese characters segmented from the license plates, which is based on the connected region feature and standard deviation feature extracted from sample corresponding to each template. Experimental results show that the proposed method can be implemented efficiently.

Handwriting is the most effective way by which civilized people speaks. Devanagari is the basic Script widely used all over India. Many Indian languages like Hindi, Marathi, Rajasthani are based on Devanagari Script. In the proposed work multistage approach i.e. an artificial neural network based classifier and statistical and structural method based feature extraction method has been employed for the recognition of the script. Optical isolated Marathi words are taken as an input image from the scanner. An input image is preprocessed and segmented. The key step is feature extraction, features are extracted in terms of various structural and statistical features like End points, middle bar, loop, end bar, aspect ratio etc. Feature vector is applied to Self organizing map (SOM) which is one of the classifier of an artificial neural Network.SOM is trained for such 3000 different characters collected from 500 persons. The characters are classified into three different classes. The proposed classifier attains 98% - 99% accuracy except special characters.

Saliency detection is an important research topic in computer vision. The traditional methods compute image saliency map, then salient segmentation is based on the corresponding saliency map. Unfortunately, overall performance of this method is poor due to the reason of losing some fine details and spatial information within image. This paper presents a new framework to overcome the drawback, named FDSRDS(Framework for Directly Salient Region Detection and Segmentation based on graph methods). Under FDSRSD, firstly, we get the foreground image by segmenting the original image via our extended grabcut algorithm. Mostly, the saliency region is within the foreground part. Secondly, we segment the foreground image into regions by means of graph based segmentation and nearest neighbor graph . Thirdly, we use relative weber's luminance rules to calculate every region’s luminance. Finally, we get the maximum luminance region which is the saliency region. Under FDSRSD framework, algorithms we proposed capture fine details and spatial relationships in saliency computation. We demonstrate impressive results by evaluating our method with other five state-of-the-art methods on the publicly available data set.

HUD (Head Up Display) image vibration detection requires the real-time performance. In this paper, after analyzing different algorithms for the movement estimation of the HUD, a new vibration detection algorithm is studied based on feature extraction. In the algorithm, after image smoothing, an improved Canny operator is proposed for the edge detection, then a proposed corner detection method is applied to extract the feature points on edges. The testing results prove that the studied algorithm works in real-time and with high precision.

Mixed noises are a characteristic of combined noises acting on a single carrier. Various mechanisms in recent past have been given in literature to restore images corrupted with Poisson and impulse mixed noise. This paper compares mixed noise removal techniques such as: Peer Group averaging (PGA), Vector Median Filter (VMF), Vector Direction Filter (VDF), Fuzzy Peer Group Averaging (FPGA), and Fuzzy Vector Median Filter (FVMF) on the basis of performance metrics such as Peak Signal to Noise Ratio (PSNR), Mean Absolute Error (MAE), Mean Square Error (MSE) and time complexity. The image size and the noise density is varied so as record these performance metrics. All the above mentioned techniques were implemented in MATLAB-11. The simulation and result shows that FVMF introduces blurring of edged but provide an output of highest PSNR value, especially for large sized images. However, for smaller images PGA provides best results of PSNR and hence a good quality of de-noised image. Also it is observed that with increase in image size the quality of the resulting image improves as the value of PSNR also increases but on increasing the impulse noise density with constant image size the image quality decreases with a constant decrease in the PSNR value.

The automatic handwriting identification is a hot topic in pattern recognition that it has been extensively studied in many languages. A Uyghur handwriting identification technique based on the stroke statistical features is proposed in this paper. Firstly, the handwriting image is preprocessed taking modified methods of grid line removal, noise reduction and thinning. Then novel stroke statistical features are extracted based on the structural character and writing styles of Uyghur handwriting. And this approach respectively achieves a top 1 and top 2 identification rates of 98.66% and 99.78% on the Uyghur handwriting data set from 224 different people. Finally, Comparison analysis of different stroke length and distance measurement method has been conducted through three different kinds of experiments, the optimal stroke length and distance measurement method is determined, and its effectiveness and stability are tested. The stroke statistical features can capture the structural character and writing style of Uyghur handwriting efficiently, and it is suitable for any languages theoretically.

Image segmentation is a key step of oil spills detection in SAR images. For the problem that the traditional multi-spectral clustering algorithm with the features extraction by GLCM (Gray-Level Co-occurrence Matrix) has such limitations as direction sensitivities and difficulties in selecting the best feature combination etc., this paper proposes a multi-scale segmentation method of oil spills in SAR images based on JSEG and spectral clustering. Multi-scale J-images are used to extract the multi-features and the Laplace matrix is clustered by the K-means method. Finally, a decision-level fusion strategy is used to fuse the segmentation results from different scales. Two sets of experiments show that, compared to the traditional spectral clustering methods based on the gray feature and multi-textual features, the proposed method has higher accuracy and stronger robustness.