In the thermoelastic stress analysis, stress distribution is measured by lock-in infrared thermography, which correlates temperature change due to the thermoelastic effect with reference loading signal. Loading signal from external source, such as load-cell, strain gage or displacement gage, is usually employed as a reference signal in the conventional lock-in technique. In this study, a self-reference lock-in infrared thermography was newly developed, in which a reference signal was constructed by using the same sequential data on thermoelastic temperature change. Temperature change in a region of interest was correlated with that in a remote area for reference signal construction. The lock-in algorithm based on the least squares method was employed for signal processing under random loading. It enabled us to measure the distribution of relative intensity of applied stress under random loading without using any external loading signal. Proposed self-reference lock-in thermography was applied for crack identification based on the detection of significant thermoelastic temperature change due to the singular stress field in the vicinity of crack tips. It was found that significant temperature change was observed at the crack tip in the self-reference lock-in thermal image, demonstrating the feasibility of the proposed technique