In digital image correlation measurements, the speckle area is manually selected before the correlation calculation is performed to define the matching area. With the development of industrial automation, facing the complex shape of the speckle area and the need to measure a large number of speckle images, it is crucial to find an automatic area extraction method. According to the characteristics of speckles and by comparing various conventional edge detection methods, a decision function based on second-order gradient entropy is proposed for automatically detecting speckle regions in images. By analyzing different speckle images, the optimal sub-region entropy size interval and the adaptive threshold interval in different speckle patterns were determined and the automatic extraction of the speckle region were completed by using connected region segmentation. The method was verified by using the actual speckle pattern. The experimental results show that when the entropy size of the subregion is more than 10 pixel, the decision function is sensitive to the speckle area. When the adaptive threshold value is within the range of Q-1.25 to Q, the speckle area and the background area can be effectively separated. The automatic extraction of a speckle area can be completed and the selection of speckle area before correlation calculation is achieved.