Filter-array spectrum sensors have been a promising structure that can be used to realize miniature spectrometers or spectrometers on-a-chip. By using multiple spectrum sensors with different spectral responses, the spectrum of a measurement object can be characterized. However, due to the low-cost and miniature design of the input optic interfaces, the intensity of the input light shining onto the imager of the sensor array may not be uniform. This unmodeled input nonuniformity could lead to a severe distortion in the spectrum measurement. Although the input nonuniformity can be alleviated by introducing dedicated input optic interface, we are interested in tackling this issue from an algorithmic perspective because 1) dedicated optics could still render 5%-10% intensity variation, and 2) the cost of computation power in electronics is potentially much lower than the cost of optics nowadays. Accordingly, we propose an iterative blind correction algorithm to solve the input light nonuniformity issue. The algorithm is based on the assumption that variation of input light intensity shall change smoothly, and hence requires no additional information. With the proposed iterative blind correction algorithm, significant improvement on the quality of spectrum reconstruction is obtained in both simulation and experimental studies.