Davut Hanbay

Continuous rotation invariant features for gradient-based texture classification

Four highly discriminative and continuous rotation invariant methods are proposed.We use the Hessian matrix and Gaussian derivative filters.Verified on the CUReT, KTH-TIPS, KTH-TIPS2-a, UIUC and Brodatz texture datasets. Extracting rotation invariant features is a valuable technique for the effective classification of rotation invariant texture. The Histograms of Oriented Gradients (HOG) algorithm has been proved to be theoretically simple, and has been applied in many areas. Also, the co-occurrence HOG (CoHOG) algorithm provides a unified description including both statistical and differential properties of a texture patch. However, HOG and CoHOG have some shortcomings: they discard some important texture information and are not invariant to rotation. In this paper, based on the original HOG and CoHOG algorithms, four novel feature extraction methods are proposed. The first method uses Gaussian derivative filters named GDF-HOG. The second and the third methods use eigenvalues of the Hessian matrix named Eig(Hess)-HOG and Eig(Hess)-CoHOG, respectively. The fourth method exploits the Gaussian and means curvatures to calculate curvatures of the image surface named GM-CoHOG. We have empirically shown that the proposed novel extended HOG and CoHOG methods provide useful information for rotation invariance. The classification results are compared with original HOG and CoHOG algorithms methods on the CUReT, KTH-TIPS, KTH-TIPS2-a and UIUC datasets show that proposed four methods achieve best classification result on all datasets. In addition, we make a comparison with several well-known descriptors. The experiments of rotation invariant analysis are carried out on the Brodatz dataset, and promising results are obtained from those experiments.

Principal curvatures based rotation invariant algorithms for efficient texture classification

The histograms of oriented gradients (HOG) and co-occurrence HOG (CoHOG) algorithms are simple and intuitive descriptors. However, the HOG and CoHOG algorithms based on gradient computation still have some shortcomings: they ignore meaningful textural properties and are unstable to noise. In this paper, two new efficient HOG and CoHOG methods are proposed. The proposed algorithms are based on the Gaussian derivative filters, and the feature vectors are obtained by means of principal curvatures. The feature vectors are rotation invariant by means of the rotation invariance characteristic of principal curvatures (i.e. eigenvalues). The experimental results on the CUReT, KTH-TIPS, KTH-TIPS2-a, UIUC, Brodatz album, Kylberg and Xu datasets confirm that the developed algorithms have higher classification rates than state-of-the-art texture classification methods. The classification results also demonstrate that the developed algorithms are more stable to noise and rotation than the original HOG and CoHOG algorithms. Two continuous rotation invariant descriptors are proposed.The proposed descriptors are based on principal curvatures which are rotation invariant.The experimental results show the power of the methods particularly in extremely noisy conditions.Our approaches give high classification performance on seven texture databases.

Classification of the grape varieties based on leaf recognition by using SVM classifier

In this paper, to classify the grape tree species, the extracted features from leaf images are classified using a multi-class support vector machines. Feature extraction stage, the grape leafs are calculated by using 9 different features. Image processing stage involves gray tone dial, median filtering, contrast, thresh holding and morphological-logical processes. In the classification stage, the obtained properties with the help of multi-class support vector machines (MCSVM) is performed classification process. In the testing phase, by applying the different leaf images is calculated the performance of model. In this study, MATLAB software was used. At the end of the test was determined the total success rate of 90.7%.

Classification of breast masses in mammogram images using KNN

Breast cancer is one of the most deadly diseases for women. Mammogram is very important imaging technique used diagnosis in early stages of breast cancer. In this study, a decision support system which helps experts to examine mammogram images in the fight against breast cancer is developed. In this study, firstly several preprocesses are applied to mammogram to make image clear and segmentation of mass is provided with an appropriate threshold value. After the segmentation processes, features of the tumor mass are obtained. The obtained features are classified as normal, benign or malignant using kNN (k-nearest neighbours) classifiers. In this study, its have been were shown that, effect of kurtosis, skewness and wavelet energy features on classification performance is shown. As a result, it has been seen that, these features improve the classification performance.

A novel texture classification method based on Hessian matrix and principal curvatures

In this study, in order to obtain similar effect with conventional gradient operation and extract more robust feature for texture, we use the principal curvature informations instead of the gradient calculation. Through this methods, sharp and important informations about the texture images were obtained by analyzing images of the second order. Considering the classification results obtained, it is shown that the proposed method improve the performance of original CoHOG and HOG feature extraction methods. As a result of experiments on datasets with different characteristics, it is seen that, the proposed method has higher classification performance.

Realization of a Hybrid Face Detecting and Verifying System

Face recognition is a popular subject in computer vision and objects recognition area because of each person has unique facial features. In this paper, the realization of a hybrid system for face detecting and verifying was presented. Gabor wavelet transform was used to extract facial features of individuals from images. An Artificial neural network was used to classify faces by using obtained features. Phase correlation method was used for face verifying. A MATLAB Graphical user interface was designed by combining these systems for realizing proposed hybrid system, after filtering and scanning methods, the obtained face areas demonstrate within an outline. Phase correlation methods were used to accelerate the searching process. The performance of the proposed system was tested on different image database. It was understood that the proposed method works with high accuracy but is slow when considered as the whole process. General Terms Face Recognition, Gabor wavelets, Phase Correlation, Artificial Neural Networks

Classification of EMG signals by LRF-ELM

Electromyogram (EMG) signal can be defined as the electrical activity of muscles cells. It is commonly used in motion recognition, treatment of neuromuscular disorders and prosthetic hand control. In this study, classification of EMG signals obtained from 6 different hand shapes of holding object was proposed. At first Short Time Fourier Transform of the EMG signal were evaluated to obtain their Time-Frekans representation. After than these T-F images were segmented and their mean values were evaluated to reduce the dimension of the images. Local Receptive Fields based Extreme Learning Machines (ELM-LRF) used to classification of these hand shapes of holding object. Evaluated accuracy is 94.12 %.

Mean shift based object tracking supported with adaptive Kalman filter

In this paper, mean shift algorithm and adaptive Kalman filter have been both utilized to realize object tracking in video sequences. Mean shift algorithm cannot give good results when the position of the tracked object is changed rapidly between sequential frames or the tracked object is occluded. In this paper, the first position of the tracked object is predicted by Kalman filter then mean shift algorithm starts to seek the object in this position. Bhattacharyya coefficient which is obtained from mean shift algorithm, is used to instantly update Kalman filters error covariance matrix and determine whether object is occluded or not. Experimental results demonstrate that the proposed method has been more efficient technique as compared to standard mean shift algorithm in case of occlusion and fast object tracking.

Computer-aided tumor detection system using brain MR images

In todays technology, computer assisted detection applications have managed to make great contributions to the field of medicine. Computer assisted detection systems aim to help radiologist about mass detection by using image processing systems. In this study, its aimed to carry out mass detection process on the images of 3D brain MRI (Magnetic Resonance Imaging). The steps followed in this study are the stage of pre-processing the stage of segmentation, identification of the areas of interests and identification of tumor. As a result of processings in the stages of preprocessing and segmentation, obtained areas of interests are labelled and attributes of these areas of interests are extracted during the stage of attributes extraction and in the last stage, the areas of interests are identified as whether they are mass or not according to these attributes. With this method applied on 845 number of magnetic resonance image sections belonging to 13 patients, it has been achieved classification success with 86.39%.

Determining noise performance of co-occurrence GMuLBP on object detection task

Object detection is currently one of the most actively researched areas of computer vision, image processing and analysis. Image co-occurrence has shown significant performance on object detection task because it considers the characteristic of objects and spatial relationship between them simultaneously. CoHOG has achieved great success on different object detection tasks, especially human detection. Whereas, CoHOG is sensitive to noise and it does not consider gradient magnitude which significantly effects the object detection accuracy. To overcome these disadvantages the CoGMuLBP was proposed. CoGMuLBP uses a new statistical orientation assignment method based on uniform LBP instead of using the common gradient orientation. In this study, detection accuracies of CoGMuLBP and CoHOG are calculated on three different datasets with NN classifier. In addition, to evaluate the noise performance of the methods, gaussian noises were added to test images and performances were recalculated. Numerical experiments performed on three different datasets show that 1) CoGMuLBP has higher detection accuracy than CoHOG; 2) using uniform LBP based gradient orientation improves detection accuracy; and 3) CoGMuLBP is more robust to gaussian noise and illumination changes. These results provide the effectiveness of CoGMuLBP for object detection.