Aiming at the problem of too much perceptual distortion and high error rates caused by high image source compression rate existing in steganography technology of information security field, an image steganography based on Karhunen-Loeve transform (KLT) optimization technique is designed. The covariance matrix and cluster mean value are obtained with iterative clustering algorithm. It divides images by adjusting the correlation values. The image data are compressed by introducing KLT algorithm. The least significant bit is adopted to hide data instead of the encrypting data. Image informations are hidden effectively by using inverse linear transformation operations and the original pixel matrix when extracting informations. Results show that image extracted by the method has higher capacity, larger signal-to-noise ratio and smaller image data distortion compared with common algorithms.