Described is an image compression technology by which a coefficient representative of a chrominance component corresponding to an (x, y) coordinate pair is coded into a symbol, along with plurality of context events associated with that symbol, to facilitate more efficient context-based arithmetic coding. For example, four context events may be generated, including one context event based on the corresponding (x, y) coefficient value in the luminance component, and three context events based on other luminance coefficients that have adjacent horizontal, vertical and diagonal relationships with the corresponding luminance component's coefficient, respectively. In one example implementation, coding the chrominance coefficient and computing the plurality of context events occurs in a zero coding pass of an entropy coder. A sign coding pass and a magnitude refinement pass may be performed to obtain further symbols for the context-based arithmetic coding.