Summary: This paper proposes a novel approach to content-based Watermarking for image authentication that is based on Independent Component Analysis (ICA). In the scheme proposed here, ICA is applied to blocks of the host image and the resulting mixing matrix represents the features of the image blocks. Frobenius norm of the mixing matrix is adopted as the content-based feature. This is embedded as the watermark in a mid frequency DCT coefficient of the block. This authentication technique is robust against incidental image processing operations, but detects malicious tampering and correctly locates the tampered regions.