We propose a general approach towards feature extraction for identifying sonar targets based on their composition and geometry. The key idea is to discover the geometric connections between braid-like features within acoustic color topography that includes magnitude and phase information. Specifically, we characterize each target as a graph of intersecting braided features, detected across the complex-valued sonar spectrogram. This enables us to create a dictionary of encoded features, each code representing the braided features of a known target. The acoustic color features of an unknown target can then be matched using a variety of encoding distances to the nearest match within the dictionary. The goal of this work is to report the key intuition behind work in progress and therefore, should be read as a position paper rather than a comprehensive report on the full idea, which is currently being validated over experimental field data.