Garima Vyas

Automatic mood detection of indian music using mfccs and k-means algorithm

This paper proposes a method of identifying the mood underlying a piece of music by extracting suitable and robust features from music clip. To recognize the mood, K-means clustering and global thresholding was used. Three features were amalgamated to decide the mood tag of the musical piece. Mel frequency cepstral coefficients, frame energy and peak difference are the features of interest. These features were used for clustering and further achieving silhouette plot which formed the basis of deciding the limits of threshold for classification. Experiments were performed on a database of audio clips of various categories. The accuracy of the mood extracted is around 90% indicating that the proposed technique provides encouraging results.

Detection of malaria parasite species and life cycle stages using microscopic images of thin blood smear

Malaria is responsible for nearly 438,000 deaths worldwide in a year. A total of 214 million cases of malaria are encountered annually. The conventional method for testing malaria is through microscopy. A blood sample of the patient is spread over a glass slide, stained with Giemsa stain and examined under a microscope. It takes a few hours and a highly trained professional to visually examine the slide and give the results. It is even more difficult to detect the different types of malaria parasite and their stages by the conventional methods. The proposed method involves acquisition of the thin blood smear microscopic image at 100x magnification, pre-processing by partial contrast stretching, separation of infected cell from the image by applying k-means clustering on the a∗b component of L∗a∗b color space, feature extraction (shape and textural) of the infected cell, feature reduction using one way ANOVA and finally training the K-nearest neighbor classifier to test the images. Instead of extracting features for the entire group of erythrocytes present in the image, the algorithm only processes the infected cells increasing the speed, effectiveness and efficiency of testing. The KNN classifier is trained with 300 images to detect three lifecycle stages (trophozite, schizont and gametocyte) for each of the four species of malarial parasites (P.falciparum, P.vivax, P.malariae, and P.ovale) with an accuracy of 90.17% and sensitivity of 90.23%.

Detection of sickle cell anaemia and thalassaemia causing abnormalities in thin smear of human blood sample using image processing

About 3.2 million people suffer from sickle-cell disease. Aim of this paper is to detect sickle cell anaemia and thalassaemia. The proposed method involves acquisition of the thin blood smear microscopic images, pre-processing by applying median filter, segmentation of overlapping erythrocytes using marker-controlled watershed segmentation, applying morphological operations to enhance the image, extraction of features such as metric value, aspect ratio, radial signature and its variance, and finally training the K-nearest neighbor classifier to test the images. The algorithm processes the infected cells increasing the speed, effectiveness and efficiency of training and testing. The K-Nearest Neighbour classifier is trained with 100 images to detect three different types of distorted erythrocytes namely sickle cells, dacrocytes and elliptocytes responsible for sickle cell anaemia and thalassemia with an accuracy of 80.6% and sensitivity of 87.6%.

An automatic emotion recognizer using MFCCs and Hidden Markov Models

In this paper, the proficiency of continuous Hidden Markov Models to recognize emotions from speech signals has been investigated. Unlike the existing work which considers prosodic features for automatic emotion recognition, this work proposes the effectiveness of the phonetic features of speech particularly, Mel-Frequency Cepstral Coefficients which improves the accuracy with reduced feature set. The continuous speech emotional utterances used in this work have been taken from the SAVEE emotional corpus. The Hidden Markov Model Toolkit (HTK) version 3.4.1 was utilized for extraction of the acoustic features as well as generation of the models. Optimizing the acoustic and pre-processing parameters along with the number of states and transition probabilities of the Markov Models, the trials give us an average accuracy of 78% and highest accuracy of 91.25% for four emotions sadness, surprise, fear and disgust.

An integrated spoken language recognition system using support vector machines

An automatic Language Identification (LID) is a system designed to recognize a language from a given spoken utterance. The spoken utterances are classified according to the pre-defined set of languages. In this paper a LID system is designed for two different languages namely English and French. The classification of an audio sample is done by extracting Mel frequency cepstral coefficients (MFCCs) and putting them on support vector machines with radial basis function kernel. The proposed framework is speaker-independent. This scheme was tested on a database of multi-lingual speech samples. The language identification accuracy is found to be 92% for French and 88% for English.

Analysis of histogram based compound contrast enhancement with noise reduction method for endodontic therapy

Radiographs are essential to all phases of endodontic therapy. They inform the diagnosis and the various treatment phases and help evaluate the success or failure of treatment. Because root canal treatment relies on accurate radiographs, it is necessary to master radiographic techniques to achieve films of maximum diagnostic quality. Such mastery minimizes retaking of films and reduces the radiation exposed on patient due to which image quality is low contrast. Hence, image processing techniques are an acceptable technique that can be used to improve the quality of image to assist dentist for diagnosis. Digital dental radiograph images are often noisy, blur edges and low in contrast. We have proposed the combination of sharpening and enhancement method to overcome these problems. The impulse noise is reduced by using median filtering technique. Then filtered image is passed through the homomorphic filter to improve the image illumination. The processed dental images for root canal teeth are contrast enhanced by sharp contrast adaptive histogram equalization. The combination of median and homomorphic filter with SCLAHE is compound contrast enhancement method. The results are verified in terms of peak signal to noise ratio (PSNR) and entropy of image.

An automatic classification of bird species using audio feature extraction and support vector machines

Automatic identification of bird species based on the chirping sounds of birds was experimented using feature extraction method and classification based on support vector machines (SVMs). The proposed technique followed the extraction of cepstral features on mel scale of each audio recording from the collected standard database. Extracted mel frequency cepstral coefficients (MFCCs) formed a feature matrix. This feature matrix was then trained and tested for efficient recognition of audio events from audio test signals. 70% of the whole database was used for training purpose while the reamaining 30% for testing of samples. The classifier achieved upto 89.4% accuracy on a data set containing four species, commonly found in India.

Segmentation of Musculoskeletal Tissues with Minimal Human Intervention

Noninvasive methods of detection of diseases are very important in the medical domain. Imaging modalities such as MRI are usually employed and present the state of the art. As of now, it is very widely used in the prognosis of heart diseases where tissue distribution is taken into account. This work exhibits multi-modal MRI to enable segmenting tissues in limb, which happens to be a crucial first step in analysis.

Computer Aided Diagnosis of Cervical Cancer Using HOG Features and Multi Classifiers

Cervical cancer is very common in women, and it is the most dreaded disease. Cervical cancer if detected early can be treated successfully. Cervical cancer occurs due to the uncontrolled growth of the cells present in the cervix of the female body, and it also occurs due to the virus human papilloma virus (HPV). Pathologists diagnose cervical cancer by a screening test called Papanicolaou test or Pap smear test. The pap smear test is not always 100% accurate but it helps in early detection of cancerous cells. In this paper, a method is proposed that helps in detection and classification of the cancer using HOG feature extraction and classifying it by the help of support vector machine (SVM), k-nearest neighboring (KNN), artificial neural network (ANN). The database was collected from Air Force Command Hospital, Bengaluru. A total of 66 pap smear images were collected that are 25 normal pap smear images and 41 abnormal pap smear images. Histogram of gradient (HOG) extracts features of the region of interest in the image as it converts pixel-based representation into gradient-based representation. The classification of cervical cells—abnormal cells and normal cells—is done with the help of multi-classifier. The accuracy attained after classification is 62.12, 65.15, and 95.5% for SVM, KNN, and ANN, respectively.

A novel approach towards identification of in-flight situation based on air traffic control conversations

With the rise in Air Traffic flow across the world due to advancement in technology and developments in the field of aeronautical engineering, the cases of emergency and panic situations on flights have also emerged at an exponential rate. Every single day, we hear of emergency situations in flights like fires, birdstrikes, diversions, engine failures and emergency landings etc. Across the globe, Air Traffic Controllers are constantly monitoring the in-flight situation based on inputs from the pilots and advicing them on possible diversions or measures to take so as to avoid any danger to the lifes of passengers and the cabin crew. Though, the manual monitoring of the Air Traffic Control (ATC) Conversations is efficient and reliable, but still, it may be subject to human biass and glitches. Also, post-flight monitoring is difficult because there is no such automatic measure for arriving at the conclusion, whether the flight faced an emergency or not. Audio Signal Processing finds an interesting use in this case. By extracting some crucial features from an audio conversation, one can proceed towards the classification once the feautres seem good and efficient. The paper mainly focusses on extracting these features critical to ATC analysis. In the future, with availablity of good database, the paper can be expanded to classify the conversations using Machine Learning techniques.