The study presents a comprehensive quantitative evaluation of CNN- enhanced compressed seeing styles for image reconstruction using Orthogonal Matching Pursuit (OMP) and Compressive Slice Matching Pursuit (CoSaMP). trials are conducted on grayscale image datasets corrupted by Gaussian noise across signal- to- noise rate (SNR) situations from 5 to 30 dB. Performance is totally assessed using regularized mean square error (NMSE) and peak signal- to- noise rate (PSNR). Results demonstrate that CoSaMP constantly achieves lower NMSE than OMP, and the operation of CNNs further improves performance. At SNR = 20 dB, CNN integration increases PSNR from 15.30 dB (OMP) and 20.33 dB (CoSaMP) to 25.52 dB and 30.32 dB, independently. The minimal NMSE is attained by CoSaMP- CNN, validating the robustness and effectiveness of this mongrel approach. relative bar maps and grouped line plots illustrate statistically significant earnings in reconstruction quality, attesting that the proposed community between classical algorithms and deep literacy yields superior noise adaptability and image dedication under adverse conditions.