Her work over the past few areas spans across primarily two major areas viz. cryptanalysis and signal and image processing.

Cryptanalysis:

Stream ciphers are popularly used in encryption systems. We have considered the cryptanalysis of stream ciphers where the keystream generator is implemented using linear feedback shift registers and a nonlinear combining function. The first ciphertext only attack, in the true sense of the word, has been developed by us, where knowledge of neither the initial conditions nor the combining functions is utilized. Estimates of the cipherlength required for determining the initial conditions as well as the combining function have been provided.

An existing class of attacks for stream ciphers, termed fast correlation attacks, commonly had a pre-processing or iterative phase. We proposed a novel fast correlation attack which, apart from dispensing with this phase and thereby being fast, also performed well even where limited lengths of the ciphertext were available. We also suggested a modification to handle a situation where the combining function was unknown.

Signal and Image processing:

Among the challenges faced in the field of functional magnetic resonance imaging (fMRI), proper registration of two-dimensional and three-dimensional volumetric images in an image-data series and correct detection of activation from the images were the problems that attracted our interest. The need for registration usually arises on account of inadvertent head movement of the patient. We proposed a robust technique for motion correction based on the Least Trimmed Squares estimator. Its performance has been established through extensive, well-designed simulations. The problem of activation detection from among artifacts and other spurious efects has been attempted through wavelet based filtering which seems to be a promising approach.

Symbolic compression is a lossy compression technique, which can be used eficiently for the compression of Indian language text documents. Digital watermarking of such documents is often important for purposes of owner verification. Symbolic compression treats the document image as a binary one, thereby making the task of data hiding very dificult, as visual artifacts are easily created due to small changes in a binary image. We proposed an approach for the watermarking of such documents, exploiting the special features of symbolic compression and based on properties of the human visual system, greatly reducing the possibility of artifacts and thus leading to a robust watermarking system. This was the first time that compression and security of Indian language text documents, in an unified framework, was provided. Further, watermark retrieval may be either from the compressed document or its reconstructed form, hence supporting operations in the compressed domain which is another advantage of our approach.

Blind quality assessment of images and video is an important requirement in multimedia communication. Embedding a known watermark in the host video before transmission, enables the receiver to gauge the degradation undergone by the video, simply by extracting the watermark and observing its quality with reference to the original, known watermark. We have proposed an approach which dispenses with the need for using an extraneous watermark and creates it from the image by making use of robust features of the image. The reference watermark for comparison at the receiving end, is constructed again from the received image following the same technique. Such an approach is the ¯rst of its kind to be used for blind quality assessment.

Blind source separation of signals from an unknown mixture is a widely studied problem with different uses. We have been studying the problem for audio signal mixtures, including both speech and music using both frequency domain and time-domain approaches. Real-time as well as offline implementations are being explored. We considered the application of this in a hearing-aid system which would enable the user to choose between typically, one out of a mixture of two signals. This requires audio signal discrimination, study of which has led us to the development of some useful discriminatory features.