Reversible watermarking is a method of hiding the watermark in digital media in such a way that visually its effect is almost negligible and after extracting the watermark, digital media can be restored to its original form bit‐by‐bit. Reversible watermarking has applications in the fields that are very sensitive towards security such as defense, medical, legal matters, artwork, and so on. This paper investigates the concept of prediction error expansion in developing a fragile reversible data hiding technique for digital images. The fact that the adjacent pixels are highly correlated is exploited by the proposed technique for providing high embedding capacity along with good visual quality. The proposed technique utilizes odd columns to predict the values of even columns and embeds the watermark into even columns. Compressed location map is used to handle the overflow and underflow problem. Experimental results using various standard test images and a comparison with recent existing techniques show that the proposed technique provides high embedding capacity with better visual quality. Higher peak signal‐to‐noise ratio values indicate the effectiveness of the proposed technique. Copyright © 2016 John Wiley & Sons, Ltd.