Coding of Aerial Reconnaissance Images for Transmission over Noisy Channels.

摘要

Several picture coding techniques were applied to aerial reconnaissance photos for the purpose of comparative evaluation. Both spatial and frequency domain methods were used: five of them previously known, as well as two new ones enunciated here for the first time. The comparisons reported within were subjective ones carried out by a team of five professional photo analysts trained to extract targets from aerial images. Their judgements were obtained on noise-free images as well as those coded and subjected to a simulated noisy transmission channel prior to decoding. The bit error rates used were 0, .0001, .001, and .01. The specific coding methods evaluated were: fixed zonal; threshold; four-class zonal (Chen and Smith); hybrid; Micro-Adaptive Picture sequencing (CDC); and the two new techniques: block truncation, and variable zone. Detailed theoretical discussions of each coding method are included as are listings of all specific parameter choices used in each method. Arithmetic operations count estimates and the resulting mean-square error performance for each technique are also given. (Author)