With tremendous advancement in the field of communication and the abundant increase in social networking sites, making duplicates of web pages, images and videos has become very simple and easy. With such a high inflow of data, finding a Near-Duplicate at very high speeds becomes a challenging mission. The complexity of this task increases further for a video, as both temporal and spatial information will have to be considered and computations and comparisons should be done at real time. This thesis addresses the issue of Near-Duplicate Detection, by reducing the computational complexity by the use of pair-wise pixel similarities. System of Associative Relations (SOAR) is used to encode inter-pixel relationships as a sequence of 1's, 0's and -1's which is compared with a huge database of different and Near-Duplicate videos. Instead of considering the pixel value, only the difference is encoded, thus only the gradient change in intensity is considered for computations which proves to be sufficient to uniquely detect a video and greatly reduces the complexity. This algorithm avoids operations like division and other complex mathematical operations and mostly uses exclusive OR's and additions which can easily be implemented in hardware using simple 'XOR' gates and 'AND' gates. The obtained results are compared with the results obtained from a similar method based on spatio-temporal signature, and a better precision and lower false positive rates are seen.