Classification of Hyperspectral Image using SVM
Post-Processing for Shape Preserving Filter and PCA

This project is based on experimentation to preserve shapes of the natural classes in a hyperspectral image post-classification of the image using SVM. The classifier classifies the vegetation types present in the hyperspectral image and then estimates the crop types present in the image. The result has been found very effective in characterizing significant spectral and spatial structures of objects in a scene.

Dark Channel Processing for Medical
Image Enhancement

The images obtained through biomedical instruments are not always satisfactory. They are often blurred or vague. This project concentrated on image processing system which enhances these biomedical images. Satisfactory results were obtained which showed clearer biomedical images. Further SSIM values and correlation factor are calculated to compare the results. The results were tested for different images obtained by medical instruments.

Inspection of Vehicles using Machine Vision

The inspection process these days in railways is mostly manual hardships which result in less accuracy and efficiency. To minimize errors and increase the work efficiency machine vision systems are being developed. This system uses advanced cameras stationed trackside to capture images of the train as it passes by. This inspection system uses deep learning algorithms - UNet, to capture multiple images and learn about faults in the spring. Furthermore, to increase the efficiency of the algorithm, a basic optimizer called RMSProp was modified. This project's paper is not published as of yet.

Dehazing of Aerial Images using Dark Channel
and Gamma Correction

An effective method to improve the contrast of hazed images by computing its dark channel image, calculating the atmospheric light, recovering the scene radiance and refining it by gamma correction. On testing the algorithm on real aerial images, we obtain significant results. This method is applicable to colored and gray-scale images. The experimental results show the SSIM index values and correlation factor nearly equal to one, after executing the algorithm on hazed images.

Automated Analysis of Dislocations present
in a Semi-conductor Material

With the help of image processing in MATLAB software, the number of defects, enhancement of image, and study of dislocation density are analyzed which helped in a better understanding of the material. HgCdTe material is widely used as an infrared radiation detector and while fabricating it, etching introduces many defects. MATLAB’s built-in development environment really helped in doing the image processing. Calculation of the defects in the material after etching helps in observing the effect of etchants on the material.

Self-Parking Car prototype

The prototype focuses on self-driving with parking in an environment created by Raspberry pi. The car is planned to be made smarter by data handling and machine learning. The results showed how self-parking can be implemented on regular cars and an ample amount of time and money can be saved by parallel parking systems.