Computational burden due to high dimensionality of Hyperspectral images is an obstacle in efficient analysis and
processing of Hyperspectral images. In this paper, we use Kernel Independent Component Analysis (KICA) for
dimensionality reduction of Hyperspectraql images based on band selection. Commonly used ICA and PCA based
dimensionality reduction methods do not consider non linear transformations and assumes that data has non-gaussian
distribution. When the relation of source signals (pure materials) and observed Hyperspectral images is nonlinear then
these methods drop a lot of information during dimensionality reduction process. Recent research shows that kernel-based
methods are effective in nonlinear transformations. KICA is robust technique of blind source separation and can
even work on near-gaussina data. We use Kernel Independent Component Analysis (KICA) for the selection of
minimum number of bands that contain maximum information for detection in Hyperspectral images. The reduction of
bands is basd on the evaluation of weight matrix generated by KICA. From the selected lower number of bands, we
generate a new spectral image with reduced dimension and use it for hyperspectral image analysis. We use this technique
as preprocessing step in detection and classification of poultry skin tumors. The hyperspectral iamge samples of chicken
tumors used contain 65 spectral bands of fluorescence in the visible region of the spectrum. Experimental results show
that KICA based band selection has high accuracy than that of fastICA based band selection for dimensionality reduction
and analysis for Hyperspectral images.
This paper presents a new method for detecting poultry skin tumors based on serial feature fusion in hyperspectral
images. First, some transform methods, including principal component analysis, discrete wavelet transform and band
ratio method, are used to generate largely independent datasets in the hyperspectral fluorescence images. Then, the
kernel discriminant analysis is utilized to extract features from each represented dataset for the purpose of classification;
another set of features are extracted from hyperspectral reflectance images by using kernel discriminant analysis. Finally,
new fused features are made by combining aforementioned features. The experimental result based on the proposed
method shows the better performance in detecting tumors compared with previous works.
This paper presents hyperspectral fluorescence imaging and a support vector machine for detecting skin tumors. Skin cancers may not be visually obvious since the visual signature appears as shape distortion rather than discoloration. As a definitive test for cancer diagnosis, skin biopsy requires both trained professionals and significant waiting time. Hyperspectral fluorescence imaging offers an instant, non-invasive diagnostic procedure based on the analysis of the spectral signatures of skin tissue. A hyperspectral image contains spatial information measured at a sequence of individual wavelength across a sufficiently broad spectral band at high-resolution spectrum. Fluorescence is a phenomenon where light is absorbed at a given wavelength and then is normally followed by the emission of light at a longer wavelength. Fluorescence generated by the skin tissue is collected and analyzed to determine whether cancer exists. Oak Ridge National Laboratory developed an endoscopic hyperspectral imaging system capable of fluorescence imaging for skin cancer detection. This hyperspectral imaging system captures hyperspectral images of 21 spectral bands of wavelength ranging from 440 nm to 640 nm. Each band image is spatially co-registered to eliminate the spectral offset caused during the image capture procedure. Image smoothing by means of a local spatial filter with Gaussian kernel increases the classification accuracy and reduces false positives. Experiments show that the SVM classification with spatial filtering achieves high skin tumor detection accuracies.
We present the principles and applications of our dual-modality fluorescence and reflectance hyperspectral imaging (DMHSI) system. In this paper we report on background work done using laser induced fluorescence (LIF) by the group in the early detection of esophageal cancer. We then demonstrate the capabilities of our new DMHSI system. The system consists of a laser, endoscope, AOTF, and two cameras coupled with optics and electronics. Preliminary results, performed on mouse tissue, show that the system can delineate normal and malignant tissue regions in real-time.
Hyperspectral fluorescence images reveal useful information for detecting skin tumor on poultry carcasses. In this paper, a hyperspectral fluorescence imaging system with fuzzy interference scheme is presented for detecting skin tumors on poultry carcasses. Image samples are obtained from a hyperspectral fluorescence imaging system for 65 spectral bands whose wavelength is ranged from 425(nm) to 711(nm). The approximation component of the level-1 decomposition of discrete wavelet transform is used for processing to reduce a large amount of hyperspectral image data. Features are computed from two spectral bands corresponding to the two peaks of relative fluorescence intensity. A fuzzy interference system with a small number of fuzzy rules and Gaussian membership functions successfully detects skin tumors on poultry carcasses.
This paper presents classification of difference image blocks between the two successive image frames for video data compression. Difference blocks are classified to several activity categories according to the image activity distribution. The classification procedure goes in two steps: activity classification and distribution classification. In the activity classification, each interframe difference image block is classified into active or not-active class according to the amount of motion contained in the block. Distribution classification further classifies active image blocks to four activity categories, vertical, horizontal, diagonal, and uniform activities, based on the activity distribution measured by the edge feature vector in the discrete cosine transform domain. A multiplayer feedforward neural network, trained with a small set of sample classification data, successfully classified difference image blocks according to edge feature distribution. The classification scheme improves the performance of video compression at a cost of small increase in the overhead associated with the quantizer switching.