Abstract: In order to extract the image feature automatically,a novel data field-based method for sparse binary image was proposed from the point of view of the physics-like field theory．First,the method constructs a map from grayscale space to potential space by producing the data field for a given binary image．Next,it calculates the potential value and the principal direction for each pixel with non-zero value by scanning its 8-connected regions,and then obtains the potential matrix and the direction angle matrix．Finally,it generates the feature vectors and its corresponding visual curve after the normalization of potential value and principal direction．The proposed method solves the issue on image feature extraction using data field,and it can keep a balance between the locality of image grayscale space and the globality of potential space in data field．The quantitative and qualitative experiments with the handwritten digital images indicate that the proposed method yields accurate and robust feature extraction results,and is reasonable and effective.

Key words: Data field,Cognitive physics,Image feature,Binary image