Shape　from　focus　(SFF)　is　a　widely　used　approach　for　reconstructing　3D　shapes　from　a　sequence　of　multifocal　images.　However,　the　quality　of　this　sequence　directly　affects　the　reconstruction　accuracy.　This　study　addresses　the　challenges　posed　by　untextured　abnormal　regions　caused　by　specific　surface　characteristics,　which　significantly　compromise　reconstruction　precision.　To　improve　the　reconstruction　accuracy　at　the　data　source,　we　first　generated　an　initial　depth　map　from　the　multifocal　image　sequence,　followed　by　preprocessing　to　create　an　all-infocus　image.　This　image　then　undergoes　detection　and　segmentation　to　identify　abnormal　regions,　particularly　those　resulting　from　overexposure.　For　efficiency,　a　mask　was　applied　to　exclude　texture-rich　regions　from　the　fusion　process.　Next,　segmented　abnormal　regions　are　categorized　by　size.　Large　regions　undergo　a　rescan,　and　the　wavelet　transform　method　fuses　multiple　exposure　images　to　yield　a　reliable　data　source.　Shape　reconstruction　was　sequentially　performed　on　these　refined　data　by　incorporating　filtering　techniques　to　enhance　depth　map　quality.　Our　experiments,　conducted　on　both　synthetic　and　real　image　sequences,　validate　the　effectiveness　of　the　proposed　method.　In　synthetic　sequence　results,　the　correlation　index　exceeded　0.999,　demonstrating　high　structural　alignment.　The　proposed　approach　significantly　outperforms　existing　methods,　achieving　superior　results　in　high-precision　shape　reconstruction.