Date of Submission


Date of Award


Institute Name (Publisher)

Indian Statistical Institute

Document Type

Doctoral Thesis

Degree Name

Doctor of Philosophy

Subject Name

Computer Science


Electronics and Communication Sciences Unit (ECSU-Kolkata)


Chanda, Bhabatosh (ECSU-Kolkata; ISI)

Abstract (Summary of the Work)

An image is a recorded replication of natural scene or objects using suitable sensor and recording media. The visual quality of the recorded image may be enhanced using various types of low-level processing namely noise smoothing, contrast enhancement. The images of the same scene recorded by several sensors reveal more inforination but in their own respective ways. This advantage of multi-sensor imaging system is real- ized through the fusion of the multimodal images. One more important higher-level processing is segmentation where the image is decomposed into a set of meaningful regions ( e.g. objects and background). An image, in general, contain objeets of varying scales or size. Therefore the objects of an image should be treated according to its scale. Image processing operations in which the objects are treated as per their scales are termed as multi-scale techniques. The associated scale-space should satisfy a number of criteria namely causality, rotational invariance, edge localization. sculr- invariance, scale-calibratedness. Mathematical morphology is a special class of image processing technique, that, by and large, preserves the shape of the features in the filtered versions of the image. Morphological operations may be extended in multi- scale sense. Multi-scale morphological operations take care of both the scale and the shape of the features in an image. In this thesis we have implemented a morpho- logical scale-space, based on structural opening and closing with convex structuring element, using a stack of images of same size and refer it by morphologicul tower. Then the proposed implementation has been employed as a tool to solve a number of problems in image processing which include noise smoothing, contrast enhancement for graylevel images, contrast enhancement of color images, fusion of multimodal 21) images and segmentation of graylevel images.In noise snoothing, the features of the image at various scales corrupted with noise. are cxtracted and stacked in different levels of morphological towers. While construct- ing the smoothed image by recombining the feature images we assign progressively smaller weights to feature images at lower scales since the noise affects the features of smaller scales more. In local contrast enhancement the approach is same as that of noise smoothing except the order emphasis given to the fcature images. The con- struction of locally enhanced image involves recombining the feature images with progressively more weights to features of smaller scales.Enhancing the contrast of color images is an extension of that of graylevel images. The intensity image constructed by considering the magnitude of the color vector at pixel location is enhanced like a graylevel image. The enhanced color image is obtained by combining the enhanced intensity image with the preserved direction cosines. The hue and the saturation of the color image are preserved. In image fusion, the scale-specific features extracted from the registeredpair of im- ages are stacked in different morphological towers. At each scale we select the features which are best represented in any of the modalities. The selected features are com- bined to construct the fused image.


ProQuest Collection ID:

Control Number


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Included in

Mathematics Commons