Sahin Isik

Prediction for the solid waste composition by use of different curve fitting models: A case study

This study presents a methodology for the prediction of solid waste composition in the urban area based on a set of limited samples. The methodology was applied by a case study for Eskişehir city in Turkey. For this purpose, Municipal Solid Waste (MSW) samples were collected for one year according to socioeconomic structure of districts. MSW samples were separated mainly into five groups of: paper-cardboard, metals, glass, plastics and food wastes as manually. The 75% of the values for each group were used as train data sets and the remains were used as test sets considering to income levels and population. It was used different curve fitting models for training of data and obtained different equations (power, exponential and polynomial) from the models. These equations were used for prediction of test sets and real values and test results were compared. Prediction accuracies were determined and interpreted according to different goodness of measurement values. It was seen that the effect of income level and population on waste composition from the degree of accuracy of this model is very important.

A novel multi-scale and multi-expert edge detection method based on common vector approach

Edge detection is most popular problem in image analysis. To develop an edge detection method that has efficient computation time, sensing to noise as minimum level and extracting meaningful edges from the image, so that many crowded edge detection algorithms have emerged in this area. The different derivative operators and possible different scales are needed in order to properly determine all meaningful edges in a processed image. In this work, we have combined the edge information obatined from each operators at different scales with the comcept of common vector apprach and obtained edge segments that connected, thin and robust to the noise.

ImageHawk search engine: Content based image retrieval system

This paper describes large-scale content based image retrieval system, Image Hawk search engine. ImageHawk search engine uses 23.4 million images in its gallery. Users have two different methods to make their search: Product Quantization (PQ) and Transductive Support Vector Machine based Hashing using Binary Hierarchical Trees (TSVMH-BHT). Images are first represented with 20480-dimensional Fisher vectors and then binary codes are extracted from Fisher vectors by using these two methods. 256-bit binary codes are used for PQ and 512-bit binary codes are used for TSVMH-BHT. When a query image is given to the search engine, the system returns the most similar 100 images in 30–40 seconds based on the size of the query image. In addition we also describe our new image retrieval dataset created by using ImageCLEF 2013 and report the accuracies of some popular image retrieval methods on this dataset.

Quality measurement of fruits and vegetables with near infrared camera

The determination of the quality of some foods with traditional techniques is considered as deceptive and costly. For this reason, more objective and fair methods are preferred for the measurement of food quality. In food industry it is very common to use image processing methods that uses the color, size etc. of foods to measure the quality and to classify them. In this study, it is aimed to measure the quality of food products (vegetables and fruits) with computerized systems as easily and objectively by a near infrared spectroscopy.

A new approach for speech denoising

With increasing use of communication via digital signal speech data in real world applications, noise reduction in speech data became an important requirement. Traffic, crowd and such uncontrollable environmental parameters can be accounted for the background noise in speech data. One of the speech denoising approaches is to measure noise characteristics at the moments that speech do not exist and remove it from the entire speech data using statistical or spectral subtraction methods. In this paper, the concept of Singular Value Decomposition (SVD) is applied on spectral components of speech data, reconstructing the denoised speech data using inverse Fourier transform.

Fusion of multispectral images using a new subspace approach

Fusion of multispectral images is a major research problem in terms of surveillance, remote sensing, industrial automation, medical and defense applications. The main reason behind multispectral image fusion is that using single channel images does not meet requirements of classification, segmentation and related tasks of remote sensing applications. Therefore, a new solution is needed to combine multispectral images to get more accurate and good visualized image as well as preserving the important details behind them. With this purpose, we have a proposed a new image fusion technique by adopting the Common Vector Approach concept. Upon examining the results, one can observe that using the CVA method for image fusion promises good results when compared with Principal Component Analysis and Singular Value Decomposition.

Multispectral image fusion based on the common vector approach

Multispectral image fusion has attracted much attention in the area of computer vision based image processing for remote sensing, industrial automation, surveillance, medical and defense applications. The process carried out in image fusion is combining useful information stated on different channels related to the same scene. Since the proposed image fusion technique greatly improve the performance of image classification, segmentation and edge detection, a new solution is required to combine multispectral images in order to get more informative and good visualized one as well as preserving the important details behind them. By considering this fact, we have introduced a new image fusion approach based on the Common Vector Approach (CVA) concept. By examining the visual results, one can observe that CVA method presents good results as compared with Principal Component Analysis (PCA), Independent Component Analysis (ICA) and Singular Value Decomposition (SVD).

Weighted principal component analysis based edge linking

As a complicated and troublesome research area, the edge detection is a fundamental step in terms of some image processing tasks including segmentation, compression and registration. In this study, we present a new approach for edge linking by applying the concept of the PCA on different types of images to determine the attractive edge segments. To determine the direction by using the angle information, the PCA decomposition is carried out on the block around the processed point. Specifically, the horizontal and vertical directions are taken into account by considering the angle between the eigenvectors corresponding to the largest and smallest eigenvalues. After making some experiments on noisy and noise free images, we have observed that the proposed method is robust to noise, preserves the structure of image and extracts well-localized and straight lines.

The assessment of feature selection methods on agglutinative language for spam email detection: A special case for Turkish

In this study, the assessment of three different feature selection methods including Information Gain (IG), Gini Index (GI), and CHI square (CHI2) is made by utilizing two popular pattern classifiers, namely Artificial Neural Network (ANN) and Decision Tree (DT), on the classification of Turkish e-mails. The feature vectors are constructed by the bag-of-words feature extraction method. This paper is focused on the Turkish language since it is one of the widely used agglutinative languages all around the world. The results obviously reveal that CHI2 and GI feature selection methods are more efficacious than IG method for Turkish language.

The investigation on the effect of feature vector dimension for spam email detection with a new framework

In this study, the effect of dimension for a feature vector on the classification of Turkish e-mails as spam or legitimate is investigated. Although hundreds of experimental studies are achieved especially for English, which is a non-agglutinative language, the number of efforts for Turkish, which is one of the most popular agglutinative languages in the world, is counted something on the fingers of one hand. Therefore, a solution is sought for Turkish spam e-mail problem taking the special characteristics of Turkish e-mails into consideration. The developed spam filtering framework has four components named as morphological decomposition, feature selection, training, and test phases. A fixed-prefix stemming approach is used to extract the features of an e-mail and then the Mutual Information (MI) method is carried out as the feature selection method. The Decision Tree (DT) and Artificial Neural Network (ANN) classifiers are employed and the recognition accuracies obtained from these methods are considerably satisfactory. The highest accuracy rates are 91.08% for ANN and 87.67% for DT methods when the dimensions of feature vectors are selected as 150×5) and (75×5), respectively.