A strong correlation exists between pixels spatially located in proximity in any given image. Predictive coding exploits this correlation by trying to predict the pixel to be encoded. This predictive is made on the basis of previously transmitted pixel values in the encoded form. Once nit is done, only the prediction error is transmitted. Predictive coding is also called Differential Pulse Code Modulation (DPCM). The philosophy underlying predictive coding is to remove mutual redundancies between successive pixels and encode only the new information. Consider the image fn and let fn be the values of the reproduced image. When fn arrives a quantity fn1, an estimate of fn is predicted from the decoded samples fn-1,fn-2............i.e.

Fn=Ψ{ fn-1,fn-2.....................}

Where Ψ is the prediction rule which is set right in the beginning. From the above diagram,

En ▲ fn - fn

The coding process continues in a recursive manner. This is what Differential Pulse Code Modulation (DPCM) is all about. In this, the coder has to calculate teh replaced sequence fn. The decoder is imply the predictive loop from the coder.

Rewriting the equation as,

Fn= fn + en

Subtracting equation we get,

δ fn= fn- fn = en – en = qn

This simply means that the error in the input sequence is equal to the quantized error in en i.e. qn.