Sathit Intajag

Retinal Image Enhancement in Multi-mode Histogram

Evaluation of a retinal image is widely employed to help doctors diagnose many diseases, such as diabetes or hypertension. From acquisition process, retinal images often have low grey level contrast and dynamic range. This paper proposes histogram analysis for solving the problems of retinal image enhancement. The proposed method uses fuzzy set to enhance the images by partitioning the image histogram to multi-modes with derivative equation. Each mode of the histogram is contrasted by finding the optimal crossover point of S-function with an index of fuzziness, which is designed for contrasting a field of view in the images. Our algorithm can achieve a number of properties of an importance for contrast stretching, such as compressed the noise in background and produced a high contrast in the field of view that provide feasibility in diagnosis from a retinal image.

Speckle Noise Estimation with Generalized Gamma Distribution

Speckle noise is an inherent property of a synthetic aperture radar (SAR) image, and it generally tends to reduce the image resolution and contrast. The speckle noise estimation is an important prerequisite, whenever SAR image is used for object segmentation. Among the many methods in statistical description that have been proposed to perform the estimation, there exists a class of approaches that use a multiplicative model of speckled image formation, such as Rayleigh distribution, K-distribution, Weibull distribution etc. In this paper, generalized gamma (GG) distribution is used to estimate the noise characteristics. GG distribution is especially attractive because it contains several distributions as special cases, viz. Rayleigh, exponential, Weibull, and log-normal. The major parameter of the GG distribution is estimated according to maximum likelihood (ML) principle. The proposed method works successfully when the solution is located in the parameter space. For verifying the performance of the proposed scheme compared to the other methods, we use x2 goodness-of-fit (GOF) test

Color retinal image enhancement by Rayleigh contrast-limited adaptive histogram equalization

Retinal fundus image is important for ophthalmologist to identify and detect many vision-related diseases, such as diabetes and hypertension. From an acquisition process, retinal images often have low gray level contrast and low dynamic range. This paper proposes a method using improved nonlinear hue-saturation-intensity color model(iNHSI) to preserve color information of the retinal images. The intensity component is enhanced by Rayleigh transformation in contrast-limited adaptive histogram equalization (Rayleigh CLAHE) algorithm. This algorithm help to increase the contrast and improve the overall appearance. The proposed algorithm was tested by using standard public database for benchmarking diabetic retinopathy detection from digital image. The proposed method can preserve the original hue component unchanged; because, the hue information of the input images is important to ophthalmologist in diagnosis process.

Automatic image segmentation using fuzzy hit or miss and homogeneity index

This paper proposes an automatic image segmentation algorithm. Our hierarchical algorithm uses recursive segmentation that consists of two major steps. First, local thresholding is carried out by the fuzzy hit‐or‐miss operator, which allows dynamic separation of a grey‐scale image into two classes, based on local intensity distributions. The fuzzy hit‐or‐miss, being an operator of fuzzy mathematical morphology, plays an important role in performing the dynamic local segmentation. This operator gives a better shape description than global thresholding methods. It also retains small but significant regions in satellite images. Second, the homogeneity index is measured in each class based on the quality of normalized intra‐region uniformity. The proposed method has been tested using both synthetic and satellite images successfully; moreover, the algorithm can estimate the number of classes automatically.

Resourceful method to remove mixed Gaussian-Impulse noise in color images

It is a challenging task to suppress mixed noise in a color image. Simple fuzzy method could reduce the mixed Gaussian-Impulse noise with preservable edge and detail of image; however, the method provides some drawbacks and led to inappropriate outputs. This paper proposed a resourceful method to remove the mixed Gaussian-Impulse noise by designing the sequential cases to estimate the optimal weights in small window for filtering the noise signals. The sequential cases consisted of impulse detection, fuzzy system for initial weights, improving the weights and optimizing the weights, and finally the output pixels estimated by either alpha trimmed mean or convex hull techniques. As depicted in the experimental results, the proposed algorithm provided the best solutions when comparison with the vector median filter and the simple fuzzy method.

Evaluation of fusion techniques for Pan-sharpened THEOS imagery

Pan-sharpened is an image fusion technique, which is designed to increase the resolution of multispectral (MS) images using panchromatic (Pan) image. In this paper, we evaluate fusion techniques for Pan-sharpened THEOS (THailand Earth Observation System) images. A calibrated THEOS imagery with heterogeneous land cover types was evaluated by different fusion techniques. This paper studied both intensity-hue-saturation (IHS) transformation and multiresolution analysis (MRA) to evaluate and improve the sharpening performance. The fusion results were visually and objectively evaluated.

Retinal image enhancement by indices fuzziness

Evaluation of a retinal image is widely employed to help doctors diagnose many diseases, such as diabetes or hypertension. From acquisition process, retinal images often have low grey level contrast and dynamic range. This paper proposes indices of fuzziness for solving the problems of retinal image enhancement. Histogram of the retinal image is stretched by estimating the optimal parameters of the indices of fuzziness in S-function, which is designed for contrasting a field of view in the images. Our algorithm can achieve a number of significant properties for contrast stretching, such as compressed the noise in background and produced a high contrast in a field of view or foreground that provides feasibility in diagnosis from a retinal image.

Generalized extreme value filter to remove mixed Gaussian-impulse noise

Noise removal in image restoration is an important technique of image processing. In this paper, a new efficient approach is proposed for removing the mixed Gaussian-impulse noise in a color image. The proposed method utilizes the concept of local rank ordered absolute distances to measure similarity between a processing pixel in the small window and their neighborhood pixels in the processing block. The generalized extreme value distribution was employed to estimate weighted averages of the pixels in the processing block for filtering the mixed Gaussian-impulse noise. From the experimental results, our filter has yielded the better results in suppressing high density levels of the mixed noise in the color images than the state-of-the-art denoising methods.

Spectral Preservation of Pan-sharpening for THEOS imagery

Image fusion in remote sensing is usually called pan-sharpening, which is a useful method to synthesis a high resolution multispectral image (MS) from the combining of a high resolution panchromatic image (PAN) with a low resolution MS image. The popular fusion methods are intensity-hue-saturation (IHS)-based methods. However, the IHS-based methods have two major problems: (i) out-of-gamut due to transformation between red-green-blue (RGB) and IHS color systems and (ii) color distortion due to variation of saturation and intensity in image fusion. The proposed method studied on the relationship between intensity and saturation to preserve spectral information of the fusing THEOS images. Thus, we found a suitable color space, iHSL (improved hue-saturation-lightness) for pan-sharpening that can isolate the intensity or lightness component. The fusion method employs smooth filter-based intensity modulation technique to merge the spatial information from PAN with the intensity component from MS images. From the studied results, our method could preserve the spectral information better than the well-known IHS-based methods.

Speckle Filtering by Generalized Gamma Distribution

Synthetic aperture radar (SAR) image contains undesired artifacts in the form of a granular look, which is called speckle. Methods previously proposed for speckle noise filtering suffer from two major limitations: 1) noise attenuation is not sufficient, especially in the smooth and background areas; 2) existing methods do not preserve or enhance edges sufficiently—they only inhibit smoothing near edges. The proposed method is capable of reducing the speckle more effectively than previous methods and jointly enhancing the edge information, rather than just inhibiting smoothing. Our method employs the generalized gamma distribution (GGD) to model the SAR images and adopts maximum-likelihood method to estimate GGD’s parameters, which use for constructing the speckle filter. Finally, the proposed method is evaluated and compared to well-accepted methods through real data.