Agrawal's reversible data hiding scheme that uses mean value in encrypted image was proposed in 2016. Mean value is the top pixel of each column that is used for data extracting and image recovery. Agrawal's scheme uses a reserving room before encryption ie, it reserves space for embedding. Also, it uses secret key generated by pseudo random number generator. However, the capacity is too low compared to other schemes. Furthermore, because of using pseudo random number generator as a key, it's randomness is relatively low. This paper proposes an improved reversible data hiding scheme that eliminates the drawbacks of previous schemes. The proposed scheme uses balanced ternary for embedding, ie, not binary bit but ternary bits. Furthermore, instead of using a Pseudo Random Number Generator, a Chaotic map is used as encryptioon key to improve the robustness of encryption. Although more data is embedded, the experimental results show that the PSNR, SSIM and Corr values are nearly the same. In terms of encryption robustness, results of the number of pixel changed rate and unified averaged changed intensity value shows that it is cryptographically strong. In terms of capacity, as not binary but balanced ternary bits are embedded, the results of capacity is improved by 1.5 times. The proposed scheme is demonstrated and explanation of all procedures is given in detail.