Vipula Singh

Automatic Detection of Diabetic Retinopathy in Non- dilated RGB Retinal Fundus Images

this paper, a method for automatic detection of microaneurysms in digital eye fundus image is described. To develop an automated diabetic retinopathy screening system, a detection of dark lesions in digital fundus photographs is needed. Microaneurysms are the first clinical sign of diabetic retinopathy and they appear small red dots on retinal fundus images. The number of microaneurysms is used to indicate the severity of the disease. Early microaneurysm detection can help reduce the incidence of blindness. Here, we have discussed a method for the automatic detection of Diabetic Retinopathy (ADDR) in color fundus images. Different preprocessing, feature extraction and classification algorithms are used. The performance of the automated system is assessed based on Sensitivity and Specificity. The Sensitivity and Specificity of this approach are 94.44 % and 87.5 %, respectively.

Kirchhoff and F-K migration to focus ground penetrating radar images

Ground penetrating radar (GPR) based land mine detection has a main challenge of having an accurate image analysis method that is capable of reducing false alarms. However this image analysis depends on having sufficient spatial resolution in the backscattered signal. This paper aims at getting better resolution by applying two migration algorithms. One is by Kirchhoff’s migration using geometrical approach and other one is F-K migration algorithms with Fourier transform. The algorithms are developed using MATLAB simulations over different scenarios for stepped frequency continuous wave (SFCW) GPR.

A NEURO-WAVELET MODEL USING FUZZY VECTOR QUANTIZATION FOR EFFICIENT IMAGE COMPRESSION

Images have large data quantity. For storage and transmission of images, high efficiency image compression methods are under wide attention. In this paper, we propose a neuro- wavelet based model for image compression, which combines the advantages of wavelet transform and neural network and uses fuzzy vector quantization on hidden layer coefficients. Images are decomposed using wavelet filters into a set of sub bands with different resolution corresponding to different frequency bands. Different quantization and coding schemes are used for different sub bands based on their statistical properties. The coefficients in the lowest frequency band are compressed by differential pulse code modulation (DPCM) and the coefficients in higher frequency bands are compressed using neural network. The coefficients of the hidden layer of the neural network are further fuzzy vector quantized, which increases the compression ratio. The visual quality of the image has been increased by introducing fuzziness to vector quantization algorithm. Satisfactory reconstructed images with large compression ratios have been achieved using this scheme.

Low complexity distributed approach to Hyperspectral image compression

Application of Hyperspectral imagery has accomplished prevalence in the field of remote sensing, hence has emanated as a very progressive field for research and development for the past couple of decades. The sententious spatial and spectral correlation divulged by the Hyperspectral images is oppressed for compression. In this paper, we affirm a near lossless compression of Hyperspectral images based on distributed source coding. Slepian-Wolf theory forms the groundwork for the exertion of the distributed source coding principle. This compression methodology is enforced on to the collimated blocks of similar size. As the information content alters from block to block, a rate is attributed to each block under the attainable rate coercion. The blocks are quantized and encoded using the Slepian-Wolf coding. The side information requisite for the Slepian-Wolf encoding is forged using linear prediction model. This compression technique aims to achieve high compression performance with low complexity and is compared with the extant compression algorithms.

Lossless hyperspectral image compression based on prediction

Hyperspectral imaging technology plays an important role in the field of remote sensing applications. Hyperspectral images exhibit significant spectral correlation whose exploitation is crucial for compression. In this paper an efficient method for Hyperspectral image compression is presented based on differential prediction with very low complexity. The proposed scheme consists of a difference coder, two predictors and a Huffman codec. The processing of the pixels varies depending on their position in the image. The resulting difference between the predicted and the actual pixel values are encoded into variable-length codewords using the Huffman codebook. The performance of the proposed algorithm has been evaluated on AVIRIS images. The experimental results show that with a Compression Ratio (CR) up to 4.14, the proposed method provides a competitive performance with comparison of JPEG2000, JPEG-LS and the OCC schemes.

Through-wall imaging system using horn antennas

Through-wall Radar (TWR) is a non-destructive technique used to determine targets behind a visually obstructed area. The common problems faced by this system are false alarms and clutters present in the environment. In this paper, we discuss the design and fabrication of a pair of Ultra Wide Band (UWB) horn antennas. These antennas, along with Vector Network Analyzer (VNA) as the waveform generator is used to perform several A scans and then combine them into a single B-scan image of the targets. MATLAB is used to successfully generate a raw 2-D image of the target with a minimal error in the X,Y coordinates of the targets. The clutter in the raw image is then removed using an algorithm to obtain a filtered image. Experiments are conducted for various wall materials of different dielectric constants. The outcome of the work lays a foundation for the implementation of more complex TWR systems.

Significance of pre-processing and its impact on lossless hyperspectral image compression

ABSTRACT The primitive aspect of hyperspectral imagery is its inherent spatial and spectral correlation. This correlation is exploited by subjecting the imaging cube to compression. A new approach to accomplish lossless hyperspectral image compression has been proposed. The imaging cube is subjected to pre-processing stage prior to entropy coding. Pre-processing stage comprises band normalization, ordering of bands followed by image scanning. A new sorting technique entitled Greedy Heap Sorting is suggested. The proposed strategy yields an average compression ratio (CR) of 4.93 and average bits per pixel (bpp) of 3.08. The proficiency of the system is on par with the existing contemporary algorithms for lossless hyperspectral image compression in terms of CR, bpp and reduced complexity.

Analysis of lossy to near-lossless compression of hyperspectral imagery using prediction and multi-stage vector quantisation

ABSTRACT Compression of hyperspectral images on-board is being pursued diligently for the past decade due to the exhaustive applications in the arena of remote sensing. Compression algorithms have to be designed in the interest of using minimum number of bits to represent a pixel at the cost of maintaining the quality of the reconstructed signal. Referring to the above context, compression algorithm has been proposed based on multi-stage vector quantisation affirming a bit rate as low as possible, meanwhile preserving the quality of the signal. The spectral redundancy is well exploited in the pre-processing stage, the occurrence of which aids the functioning of vector quantisation, resulting in the lower bit rate in the subsequent stages of decorrelation. The proposed algorithm reforms the vector quantisation in terms of spectral distortion, complexity and memory requirements. The proposed strategy divulges with a peak signal to noise ratio of 82.55 dB and signal to distortion noise ratio of 92.23 dB.

Clutter reduction techniques of ground penetrating radar for detecting subsurface explosive objects

Ground Penetrating Radar (GPR) is a method based on propagation of electromagnetic waves to detect shallowly buried objects and also gives the depth of buried objects. The land mines are made out of plastic and less metallic. Without signal processing it is difficult to detect these explosive objects using GPR. The presence of clutter makes GPR-based landmine detection difficult. Most contributions in this area deal with clutter reduction techniques. There are many signal processing techniques in the literature that have been successfully applied to GPR data for the detection of subsurface explosive objects. This paper covers various Clutter reduction techniques along with results and comparison with respect to SNR (Signal to Noise Ratio) and ROC (Receiver Operating Characteristic) curves.

An efficient algorithm for peak detection on B-Scan data generated using GPR

B-Scan data generated using the Ground Penetrating Radar contains information about the apparent location of the landmines and other similar objects buried underground. The peak of the hyperbola so generated in the B-Scan represents the apparent depth and lateral distance of the landmines from the point of operation and these apparent peaks need to be migrated to their true positions using migration technique. Peak-detection is the pre-requisite to determine the apparent location of landmines or any other object buried underground and in this paper we propose an effective and efficient algorithm for the same. The proposed algorithm works for all possible conditions in a specified range given that the landmines are separated by a lateral distance of 21cm which corresponds to a gap of 9-pixels in the B-scan data. Many problems faced by the previously proposed algorithms have been overcome significantly. The point at which the landmine is present is the apex/peak of the hyperbola in the B-scan. In order to eliminate the points of non-interest, the algorithm employs filters and morphological transforms and the nearest neighbor method. Segmentation process is applied to obtain the exact apparent peaks from the transformed image of the B-scan which makes it more capable and efficient.