French PWR cooling systems include austenitic/ferritic stainless steel cast elbows. Due to the aging of these components, there is an increasing need for monitoring possible fabrication flaws near the elbow inner and outer surfaces. Various non destructive techniques are being evaluated by the EdF (French Electricity Board) for defect detection and characterization. One of these, the eddy current (EC) technique, is usually very efficient for the detection of flaws near the surface. However, large local variations in the ferrite ratio induce high structural noise thus reducing the EC detection efficiency. Image restoration using the Groshong parametric model is proposed as a means of improving the signal-to-noise ratio (SNR) and defect characterization.