This study proposes a novel algorithm for luminal strain reconstruction from sparse irregularly sampled strain measurements. It is based on the normalized convolution (NC) algorithm. The novel extension comprises the multilevel scheme, which takes into account the variable sampling density of the available strain measurements during the cardiac cycle. The proposed algorithm was applied to restore luminal strain values in intravascular ultrasound (IVUS) palpography. The procedure of reconstructing and averaging the strain values acquired during one cardiac cycle forms a technique, coined as reconstructive compounding. The accuracy of strain reconstruction was initially tested on the luminal strain map, computed from 3 in vivo IVUS pullbacks. The high quality of strain restoration was observed after systematically removing up to 90% of the initial elastographic measurements. The restored distributions accurately reproduced the original strain patterns and the error did not exceed 5%. The experimental validation of the reconstructed compounding technique was performed on 8 in vivo IVUS pullbacks. It demonstrated that the relative decrease in number of invalid strain estimates amounts to 92.05 +/- 6.03% and 99.17 +/- 0.92% for the traditional and reconstructive strain compounding schemes, respectively. In conclusion, implementation of the reconstructive compounding scheme boosts the diagnostic value of IVUS palpography.