Implementation of Lattice Method for Video Compression

Abstract

In multimedia, the computation and space requirements are main concerns about any data. The lossy techniques are used when no finer details are needed, like in human interaction systems. There are some areas, where we need exact actual file at the decoder side, for which we need lossless compression of any type of multimedia data, like speech, image or video. We can classify the complete action in three parts, compression, transmission, and decompression of a file. Our goal is to compress the file, so as to store and transmit it in minimum space and bandwidth requirements. The multimedia file usually follows a particular sequence of data values, and which can be, nearly, estimated by looking at previous data values of the file. The lattice method uses this concept to predict the value of unknown data point, by scanning previous data points, and then the difference between the predicted value and actual value is logged in to error file. This error file contains very low entropy as compared to the actual file, which helps us in compressing the file. The lattice method for speech compression follows single dimensional pattern, in which, it predicts next value, using lattice method computation, which is then compared with next actual value, and the result is stored as error value. The lattice method for image uses two dimensional model for prediction, using either quarter plane or half plane model, and then the predicted value is compared with the actual one, as done in speech compression. The half plane model provides better accuracy for predicting the next data value. The proposed lattice method for video compression uses present frame and previous frame, to predict the data value in present frames. The predictor order considered in this method is 3, which is as first in previous motion compensated frame, second and third in present frames north and east pixel values. This method is implemented in 2 steps, first, the temporal correlation and then the spatial correlation is cancelled. The lattice method provides the entropy reduction for around 40% (2.5:1), but the computation complexity is very low, with respect to other algorithms. Also it works equally well with all videos from slow to fast moving contents. The decoder is also equally simple. As compared to adaptive lossless video compression, motion compensated lossless video coding, and other methods the proposed method provides better results in terms of entropy, with almost same computation involvements. Our methods performs better than switched predictive coding method for medium and fast moving videos, while it gives comparable results then stationary video, with very small computation.