Along with the improvement of image watermarking techniques, the necessity for effectively and comprehensively evaluating various algorithms becomes imperative. In this paper, we first propose a new categorization that fits for most of the existing watermarking algorithms that work in the wavelet domain. Then an adaptive watermarking attack for evaluating the robustness of watermarking schemes that are based on the proposed categorization is presented. This attack determines the flat regions, edges, and textures of the watermarked image and based on known features of each region the proposed attack tries to destroy the watermark information. This is done by separately manipulating the wavelet coefficients of each region such that the least visual distortion will be imposed on the attacked image. We study some recent watermarking techniques in terms of the watermark invisibility and robustness through subjective experiments and objective metrics. The average peak signal-to-noise ratio (PSNR) of the watermarked image after applying the proposed attack is more than 35 dB and the average normalized cross correlation (NC) for the extracted watermark is lower than 0.4, so theoretically, the watermark is not detectable. Moreover, the proposed method does not consider any knowledge about the underlying watermarking algorithm.