Phase decorrelation, caused by changes in the surface scattering properties between two radar acquisitions, is a major limiting factor for interferometric synthetic aperture radar (InSAR) surface deformation analysis over vegetated terrain. Persistent Scatterer (PS) techniques have been developed to identify high-quality radar pixels suffering from minimal decorrelation artifacts. However, existing PS selection algorithms are often based on the statistics of InSAR amplitude and phase measurements at each individual radar pixel, and scattering signal models that take into account the phase correlation of nearby PS pixels have not been fully developed. Here, we present a new PS selection algorithm based on the similarity of phase observations between nearby radar pixels. We used this algorithm to analyze 25 C-band Envisat SAR scenes acquired over the San Luis Valley, Colorado, and 93 C-band Sentinel-1 SAR scenes acquired over the Greater Houston area, Texas. At both the test sites, the presence of dense vegetation leads to severe phase decorrelation artifacts even in some interferograms with short temporal baselines. Our algorithm can reduce the number of false positive and false negative PS pixels identified from an existing PS identification algorithm. The improved PS identification accuracy allows us to substantially increase the total number of high-quality interferograms that are suitable for time series analysis. We reconstructed spatially coherent InSAR phase observations through an interpolation between PS pixels, and recovered subtle deformation signals that are otherwise undetectable. In both the cases, the superior performance of our PS processing strategy was demonstrated using a large number of independent ground-truth data.