Accurate classification of images is essential for the analysis of mammograms in computer aided diagnosis of breast cancer. We propose a new approach to classify mammogram images based on fractal features. Given a mammogram image, we first eliminate all the artifacts and extract the salient features such as Fractal Dimension (FD) and Fractal Signature (FS). These features provide good descriptive values of the region. Second, a trainable multilayer feed forward neural network has been designed for the classification purposes and we compared the classification test results with K-Means. The result reveals that the proposed approach can classify with a good performance rate of 98%.

The computational complexity of fractal image compression is mainly because of the huge number of comparisons required to find a matching domain block corresponding to the range blocks within the image. Various schemes have been presented by researchers for domain classification which can lead to significant reduction in the time spent for range-domain matching. All the schemes propose to first separate domains into different classes and then select the appropriate class for matching with selected range block. Here, we propose a dynamic classification scheme based on local fractal dimensions. The method can be experimented with other features of image blocks measured locally. In this work we have investigated the computational efficiency of multi-way search trees for storing domain information. The domains can be listed in a B+ tree ordered on one or more selected local features of each domain.

Fractal image compression is attractive except for its high encoding time requirements.The image is encoded as a set of contractive affine transformations.The image is partitioned into non-overlapping range blocks, and a best matching domain block larger than the range block is identified.There are many attempts on improving the encoding time by reducing the size of search pool for range-domain matching.But these methods are attempting to prepare a static domain pool that remains unchanged throughout the encoding process.This paper proposes dynamic preparation of separate domain pool for each range block.This will result in significant reduction in the encoding time.The domain pool for a particular range block can be decided based upon a parametric value.Here we use classification based on local fractal dimension.

We propose a hybrid image watermarking technique making use of the advantages of fractal theory and wavelet transforms. Generally, the strength of watermarking is set to be a fixed value for the entire image. In our work the strength of watermarking in cover image is customized according to the complexity of image portions. The image is separated into complex and smooth regions based on fractal dimension. Fractal dimension measured with differential box counting (DBC) method is used to estimate the complexity of image portions. A higher value of embedding strength is used for complex regions, whereas a smaller embedding strength for smooth regions which are more sensitive to human vision system. Results show that the method produces imperceptible watermarks with higher detector response.

Automatic animal sound classification and retrieval is very helpful for bioacoustic and audio retrieval applications. In this paper we propose a system to define and extract a set of acoustic features from all archived wild animal sound recordings that is used in subsequent feature selection, classification and retrieval tasks. The database consisted of sounds of six wild animals. The Fractal Dimension analysis based segmentation was selected due to its ability to select the right portion of signal for extracting the features. The feature vectors of the proposed algorithm consist of spectral, temporal and perceptual features of the animal vocalizations. The minimal Redundancy, Maximal Relevance (mRMR) feature selection analysis was exploited to increase the classification accuracy at a compact set of features. These features were used as the inputs of two neural networks, the k-Nearest Neighbor (kNN), the Multi-Layer Perceptron (MLP) and its fusion. The proposed system provides quite robust approach for classification and retrieval purposes, especially for the wild animal sounds.

Automatic animal sound classification and retrieval is very helpful for bioacoustic and audio retrieval applications. In this paper we propose a system to define and extract a set of acoustic features from all archived wild animal sound recordings that is used in subsequent feature selection, classification and retrieval tasks. The database consisted of sounds of six wild animals. The Fractal Dimension analysis based segmentation was selected due to its ability to select the right portion of signal for extracting the features. The feature vectors of the proposed algorithm consist of spectral, temporal and perceptual features of the animal vocalizations. The minimal Redundancy, Maximal Relevance (mRMR) feature selection analysis was exploited to increase the classification accuracy at a compact set of features. These features were used as the inputs of two neural networks, the k-Nearest Neighbor (kNN), the Multi-Layer Perceptron (MLP) and its fusion. The proposed system provides quite robust approach for classification and retrieval purposes, especially for the wild animal sounds.

This paper discusses the design and implementation of a spectral fluctuation analysis based lossy audio coding scheme. We present a simple lossy audio codec, composed of an adaptive wavelet decomposition filter, a modified psycho-acoustic model, an intra-channel de-correlation block followed by quantization and coding block. The intra-channel de-correlation block does the spectral fluctuation analysis to find the successive frames found to be similar and de-correlates it and codes. A Modified psychoacoustic model with simplified masking model was designed to fit this system. The evaluation of the system was done by using European Broadcasting Union – Sound Quality Assessment Material (EBU-SQAM) stereo wave files. Experimental results show the compression ratios achieved and subjective quality evaluation report.

This paper presents a novel method based on fractal features for the classification of mammogram images. For recognition of regions and objects in the natural scenes, there is always a need for features, which are invariant, and they provide a good set of descriptive values for the region. There are numerous methods available to estimate parameters from the images of the fractal surface. In this paper we perform mammogram image classification based on fractal dimension and fractal signature. The result has shown the potential usefulness of the fractal features for image analysis. A trainable multilayer feed forward neural network has been designed for the classification of images. Experimental results shown that the proposed system can well perform a classification rate of 98%.

This paper deals with an automatic and relatively efficient method for estimating intracranial volume from MR brain images. The proposed method consists of mainly three steps, namely skull removal, image segmentation and volume calculation. The present method uses morphological operations followed by 3D connected component labeling and image subtraction for extracting the brain mask from the original brain slices. The skull stripped images are then segmented into the three tissues namely Gray Matter, White Matter and Cerebrospinal fluid using an efficient clustering technique namely, Weighted k-means clustering algorithm followed by Expectation Maximization algorithm. Finally the volume of the segmented tissues is calculated using Cavalier's estimator of morphometric volume method and some sample results are presented. The proposed method gives reliable results for making quantitative analysis and diagnosis of tissues from Magnetic Resonance brain image slices.

The existence of numerous imaging modalities makes it possible to present different data present in different modalities together thus forming multimodal images. Component images forming multimodal images should be aligned, or registered so that all the data, coming from the different modalities, are displayed in proper locations. The term image registration is most commonly used to denote the process of alignment of images , that is of transforming them to the common coordinate system. This is done by optimizing a similarity measure between the two images. A widely used measure is Mutual Information (MI). This method requires estimating joint histogram of the two images. Experiments are presented that demonstrate the approach. The technique is intensity-based rather than feature-based. As a comparative assessment the performance based on normalized mutual information and cross correlation as metrics have also been presented.

The application of fractal geometry to musical signals and instrumental recognition system is not been widely experimented. The fractal dimension D is a very important characteristic of fractals useful for segmentation. This paper introduces an instrument identification system, which uses fractal dimension for segmentation of audio signals. This research is organized in three parts. The first part of this research investigates fractal dimension based segmentation of musical sounds for feature extraction and recognition. The second part explores extraction of the feature set from the segmented musical signal. The feature set of the proposed system includes spectral, perceptual and temporal features of the musical signal. Third part of this research describes about the neural network classifiers. The system has been experimented with kNN classifier and multi-layer perceptron classifier and the performance of these were compared. The proposed system has been trained and tested with 10 different Indian musical instruments sound samples. The sample set contains solo and duet recordings. The system has shown overall recognition rate of 89.7% for solo and 82.8 % for duet.

An improved decision-based algorithm for the restoration of gray-scale and color images that are highly corrupted by Salt-and-Pepper noise, is proposed in this paper which efficiently removes the salt and pepper noise while preserving the details. The algorithm utilizes previously processed neighboring pixel values to get better image quality than the one utilizing only the just previously processed pixel value. The proposed algorithm is faster and also produces better result than a Standard Median Filter (SMF), Adaptive Median Filters (AMF), Cascade and Recursive non-linear filters. The advantage of the proposed algorithm (PA) lies in removing only the noisy pixel either by the median value or by the mean of the previously processed neighboring pixel values. Different gray- scale and color images have been tested by using the proposed algorithm and found to produce better PSNR and SSIM values.

Multiple Classifier fusion is an efficient and widely useful method of improving system performance. The classifier fusion approach to musical instrument recognition system is not been widely experimented. This paper explores in depth a classifier combination approach for the instrument classification task, studied over a diverse classifier pool, which includes K-Nearest Neighbor, Gaussian Mixture Model and Multi-Layer Perceptron classifiers. All three classifiers were trained with the same feature space, comprised of spectral, temporal, harmonic, perceptual and statistical features. The classifier fusion has been done at decision level. We employ the Sum-based and Confidence-based integration strategies to combine three classifiers k-NN, MLP and GMM. Experiments conducted on a musical sound database containing 10 different Indian musical instruments sounds prove that the proposed classifier combination approaches outperform individual classifiers.

The paper proposes an improved fast and efficient decision-based algorithm for the restoration of images that are highly corrupted by Salt-and-Pepper noise. The new algorithm utilizes previously processed neighboring pixel values to get better image quality than the one utilizing only the just previously processed pixel value. The proposed algorithm is faster and also produces better result than a Standard Median Filter (SMF), Adaptive Median Filters (AMF), Cascade and Recursive non-linear filters. The proposed method removes only the noisy pixel either by the median value or by the mean of the previously processed neighboring pixel values. Different images have been tested by using the proposed algorithm (PA) and found to produce better PSNR and SSIM values.

The existence of numerous imaging modalities makes it possible to present different data present in different modalities together thus forming multimodal images. Component images forming multimodal images should be aligned, or registered so that all the data, coming from the different modalities, are displayed in proper locations. The term image registration is most commonly used to denote the process of alignment of images , that is of transforming them to the common coordinate system. This is done by optimizing a similarity measure between the two images. A widely used measure is Mutual Information (MI). This method requires estimating joint histogram of the two images. Experiments are presented that demonstrate the approach. The technique is intensity-based rather than feature-based. As a comparative assessment the performance based on normalized mutual information and cross correlation as metrics have also been presented.

Texture analysis plays an increasingly important role in computer vision. Since the textural properties of images appear to carry useful information for discrimination purposes, it is important to develop significant features for texture. Various texture feature extraction methods include those based on gray-level values, transforms, auto correlation etc. We have chosen the Gray Level Co occurrence Matrix (GLCM) method for extraction of feature values. Image segmentation is another important problem and occurs frequently in many image processing applications. Although, a number of algorithms exist for this purpose, methods that use the Expectation-Maximization (EM) algorithm are gaining a growing interest. The main feature of this algorithm is that it is capable of estimating the parameters of mixture distribution. This paper presents a novel unsupervised segmentation method based on EM algorithm in which the analysis is applied on vector data rather than the gray level value.

This paper deals with different techniques for registration and fusion of remote sensed images. In this work the lower spatial resolution multispectral and higher resolution panchromatic images of SPOT satellite are used. These images are registered using a registration algorithm that combines a simple yet powerful search strategy based on stochastic gradient with the similarity measure as mutual information, together with a wavelet-based multi-resolution pyramid. The algorithm is found to give sub pixel registration accuracy. The study is limited to pairs of images, which are misaligned by rotation and/or translation. The registered images are subjected to a pixel level multispectral image fusion process using wavelet transform approach. Spectral quality assessments shows that compared to other conventional image fusion techniques, this fusion process using wavelet transform keeps much of the spectral information in the merged image with respect to the original multispectral one. Finally, segmentation is performed on the fused images to validate the algorithms used for registration and fusion and the results show better accuracy for wavelet based methods than the conventional methods.

In medical applications, lossless coding methods are required since loss of any information is usually unacceptable. Integer to integer lifting operations help to implement the lossless compression methods. Medical images are sparser than other natural images, so that by exploiting the sparseness feature, it is possible to achieve maximum compression efficiency. This paper modifies the sparsity index which defines the degree of sparseness of images and it proposes a histogram sorting method which is more efficient than the conventional histogram packing methods in terms of compression ratio.