Transliteration involves transformation of one script to another based on\nphonetic similarities between the characters of two distinctive scripts. In\nthis paper, we present a novel technique for automatic transliteration of\nDevanagari script using character recognition. One of the first tasks performed\nto isolate the constituent characters is segmentation. Line segmentation\nmethodology in this manuscript discusses the case of overlapping lines.\nCharacter segmentation algorithm is designed to segment conjuncts and separate\nshadow characters. Presented shadow character segmentation scheme employs\nconnected component method to isolate the character, keeping the constituent\ncharacters intact. Statistical features namely different order moments like\narea, variance, skewness and kurtosis along with structural features of\ncharacters are employed in two phase recognition process. After recognition,\nconstituent Devanagari characters are mapped to corresponding roman alphabets\nin way that resulting roman alphabets have similar pronunciation to source\ncharacters.\n