Chatter is a self-excited vibration that affects the part quality and tool life in the machining process. This paper introduces an intelligent chatter detection method based on image features and the support vector machine. In order to reduce the background noise and highlight chatter characteristics, the average FFT is applied to identify the dominant frequency bands that divide the time-frequency image of the short-time Fourier transform into several sub-images. The non-stationary properties of the machining conditions are quantified using sub-images features. The area under the receiver operating characteristics curve ranks the extracted image features according to their separability capabilities. The support vector machine is designed to automatically classify the machining conditions and select the best feature subset based on the ranked features. The proposed method is verified by using dry micro-milling tests of steel 1040 and high classification accuracies for both the stable and unstable tests are obtained. In addition, the proposed method is compared with two additional methods using either image features from the continuous wavelet transform or time-domain features. The results present a better classification performance than the two additional methods, indicating the efficiency of the proposed method for chatter detection.