“…However, Levin et al [25] stated that these priors tend to favor blur solutions over clear ones. Afterward, various novel image priors were developed to favor clear solutions, including L 1 /L 2 prior [1], sound L 0 prior [10,26], nonzero constraint prior [27], nonconvex L 1 − αL 2 prior [15,28], and saturation-value geo-metric spatial-feature prior [16]. Leveraging the sparsity advantage of L 0 prior and an efficient optimization framework [10], a series of L 0 + X style priors have been introduced [17], including dark channel prior [4], extreme channels prior [12], and local minimal intensity prior [13], and enhanced sparse prior [14].…”