Structural health monitoring (SHM) techniques have been studied over the past few decades to detect the deficiencies affecting the performance of the structures. Detecting and localizing these deficiencies require long-term data collection from dense sensor networks which creates a challenging task for data transmission and processing. To address this problem, a comparative study of two image-based compressive sensing approaches for multiple damage localization is presented in this paper. The first methodology consists of compressive sampling from the sensor network in global and local formats. Then through statistical change point analysis on the sampled datasets, and Bayesian probability estimation, the study estimates the location of damage. The second algorithm implements compressive sensing to the subset of samples obtained from the sensor network space divided into blocks. The damage existence and location are determined by statistical hypothesis testing of Discrete Cosine Transformation (DCT) coefficients avoiding the original signal recovery. In order to evaluate the performance of both algorithms, multiple damage scenarios are simulated in steel gusset plate model. The comparison results are presented in terms of compression ratios and successful detection rates.