Tanmoy Kanti Das

A HYBRID SCHEME FOR HANDPRINTED NUMERAL RECOGNITION BASED ON A SELF-ORGANIZING NETWORK AND MLP ClASSIFIERS

This paper proposes a novel approach to automatic recognition of handprinted Bangla (an Indian script) numerals. A modified Topology Adaptive Self-Organizing Neural Network is proposed to extract a vector skeleton from a binary numeral image. Simple heuristics are considered to prune artifacts, if any, in such a skeletal shape. Certain topological and structural features like loops, junctions, positions of terminal nodes, etc. are used along with a hierarchical tree classifier to classify handwritten numerals into smaller subgroups. Multilayer perceptron (MLP) networks are then employed to uniquely classify the numerals belonging to each subgroup. The system is trained using a sample data set of 1800 numerals and we have obtained 93.26% correct recognition rate and 1.71% rejection on a separate test set of another 7760 samples. In addition, a validation set consisting of 1440 samples has been used to determine the termination of the training algorithm of the MLP networks. The proposed scheme is sufficiently robust with respect to considerable object noise.

Diverging keys in wireless sensor networks

Currently, the most popular ways of dealing with the key distribution problem in sensor networks are random predistribution schemes. For relaxed, realistic assumptions about the attacker, the key infection protocol [1] is also available. In this paper, by accepting the relaxed assumptions from [1], we propose a scheme which makes pairwise keys “drift” or diverge, which enhances security and can be used as a key distribution method. The most notable feature of this scheme is that, under some assumptions about the sensor nodes, it incurs no communication overhead at all.

Recognition of Handprinted Bangla Numerals Using Neural Network Models

This paper proposes an automatic recognition scheme for handprinted Bangla (an Indian script) numerals using neural network models. A Topology Adaptive Self Organizing Neural Network is first used to extract from a numeral pattern a skeletal shape that is represented as a graph. Certain features like loops, junctions etc. present in the graph are considered to classify a numeral into a smaller group. If the group is a singleton, the recognition is done. Otherwise, multilayer perceptron networks are used to classify different numerals uniquely. The system is trained using a sample data set of 1880 numerals and we obtained 90.56% correct recognition rate on a test set of another 3440 samples. The proposed scheme is sufficiently robust with respect to considerable object noise.

Cryptanalysis of “wavelet tree quantization” watermarking scheme

Here we study a recently proposed watermarking scheme based on “Wavelet Tree Quantization” (WTQ) which has been presented by Wang and Lin in IEEE Transactions on Image Processing, February 2004. In the scheme, the wavelet coefficients corresponding to same spatial locations are grouped together. Two such groups, selected at random, are merged to form a super tree. Some of these super trees are quantized to embed the watermark signal in the image. In the process of cryptanalysis we can identify the groups which are quantized during watermark insertion process. We then select the non-quantized groups and quantize them to remove the watermark. Experimental results show that the watermark is completely removed by this attack. The cryptanalysis falls under cipher text only jamming attack which requires a single watermarked copy.

An Asynchronous Node Replication Attack in Wireless Sensor Networks

Applications of wireless sensor network (WSN) are growing significantly, and many security protocols meant for WSN have been proposed. One of the unique problems of WSN is that the sensor nodes are not tamper resistant as the main attraction of deploying WSN is its low cost. Node replication attack exploits this weakness to launch an attack, in which cryptographic secrets from the compromised sensor nodes are used to create duplicate sensor nodes in large number. Then these sensor nodes are placed in critical locations of the WSN to mount attacks. Several protocols were proposed to defend WSN against the replication attack, and one of the promising among them is distributed detection protocol presented by Parno et al. at IEEE S&P 2005. However, we show in this paper that their distributed detection protocol is vulnerable to an asynchronous node replication attack. Further, we modify the protocol to make it secure for dynamic WSN supporting node mobility.

Analysis of the Wavelet Tree Quantization watermarking strategy and a modified robust scheme

Here we study a recently proposed watermarking scheme based on the paper “Wavelet Tree Quantization” (WTQ) by Wang and Lin (IEEE Trans Image Process 13(2):154–165, 2004). In given scheme, wavelet coefficients corresponding to the same spatial locations are grouped together. Two such groups, selected at random, constitute a supertree. Some of these supertrees are quantized to embed the watermark information in the image. In the process of cryptanalysis we first identify the groups which are quantized during the watermark insertion process. Then we select the non-quantized groups and quantize them too, to remove the watermark. Experimental results show that the watermark is completely removed by this attack. The cryptanalysis falls under the cipher text only jamming attack which requires only a single watermarked copy. Further, we suitably modify the WTQ scheme to make it robust against such cryptanalytic attacks.

Cryptanalysis of Chu's DCT based watermarking scheme

In 2003, Chu proposed an oblivious watermarking algorithm by modifying the CKLS scheme proposed by Cox, Kilian, Leighton, and Shamoon in 1997, known as the CKLS scheme. In this correspondence, we report that the modification presented by Chu is susceptible to a suitably modified attack devised by Das and Maitra in 2004. In fact, the experimental results show that Chus scheme is even weaker than the CKLS scheme in terms of our attack.

A Robust Block Oriented Watermarking Scheme in Spatial Domain

In this paper we present an invisible spatial domain watermarking technique. The technique divides the image into n small blocks and the intensity of some of these blocks are modified depending on the key, which is a secret random binary string of length m < n. Given an attacked image, the recovery process generates a bit pattern of length m and we show that from this pattern it is possible to get back the exact key using either standard correlation measure or error correcting codes. The recovery process employs window matching technique on the attacked image, which has been found to work successfully. Our method can survive common image transformations and intentional attacks both in spatial and frequency domain. Most importantly, it also survives the nonlinear geometric attacks, e.g., Stirmark. To the best of our knowledge, here we experiment with an exhaustive set of attacks which has not been provided elsewhere.

Security evaluation of generalized patchwork algorithm from cryptanalytic viewpoint

In this paper we present a cryptanalysis of the generalized patchwork algorithm under the assumption that the attacker possesses only a single copy of the watermarked image. In the scheme, watermark is inserted by modifying randomly chosen DCT values in each block of the original image. Towards the attack we first fit low degree polynomials (which minimize the mean square error) on the data available from each block of the watermarked content. Then we replace the corresponding DCT data of the attacked image by the available data from the polynomials to construct an attacked image. The technique nullifies the modification achieved during watermark embedding. Experimental results show that recovery of the watermark becomes difficult after the attack.

Cryptanalysis of Block Based Spatial Domain Watermarking Schemes

In this paper we propose a block based method of image reconstruction which acts as a generic attack against any block based watermarking scheme in spatial domain. Our reconstruction method relies on the geometric features of the image to discover and modify the inconsistencies in the pixel values introduced due to watermarking. To show its effectiveness we mount the attack on a few existing watermarking schemes. These schemes divide the image into some disjoint sets of pixels. Intensities of constituent pixels of each such set is modified to embed one bit of watermark. And the watermark is a secret binary string which comes from error correcting code. In the process of cryptanalysis we reconstruct an image from a single watermarked copy. Experimental result shows that from the attacked image, watermark extraction becomes impossible, making cryptanalysis successful.