Sparse colour and grey scale image restoration using a morphological method

In an experimental study on colour and grey scale images covering a range of corruption from 50 to 95%, good restoration was achieved using a new morphological method. The nearest good neighbour (NGN) morphological filter copies grey scale values from the `good´ pixels in a regular manner so that these pixels become seeds for the restoration during successive iterations. A complementary propagation method is also described. In this context, sparse data refers to an image in which a large fraction of the data has been replaced by impulsive noise. The impulse noise may have a high value or range or a zero or null value. Random loss of an image communication channel will result in a sparse image which may be restored by these methods. Range images or optic flow data may be processed by these methods also. On a face image which had lost 95% of its data, the restored image had a signal to noise ratio of 16 dB and all features were clearly discernable. The filter had an 8 pixel neighbourhood and took about 0.5 second per iteration to filter the image on a 386 standard PC