*Paper Organization*

The remainder of paper is organized as follows: Section 2 presents the prior superpixel creation research and its relevance to semi-dark technology. In this section we also present critical analysis of studies proposed over period of two decades and their possible applicability for semi-dark imagery. We also critically analyze two closely related studies and highlight the difference among them w.r.t SLIC++. Section 3 describes the extension hypothesis and the final detailed proposal for super-pixel segmentation of semidark images. Section 4 presents the detailed quantitative and qualitative analysis of SLIC extension against state-of-the-art algorithms validating the proposal. Section 5 discusses the applicability of the proposed algorithm in the domain of computer vision. Finally, Section 6 concludes the presented research and points out some future directions of research.

#### **2. Literature Review**

#### *2.1. Limited Semi-Dark Image Centric Research Focusing Gradient-Ascent Methods*

The gradient ascent methods are also called clustering-based methods. These methods take the input image and rasterize the image. Then, based on the local image cues such as color and spatial information the pixels are clustered iteratively. After each iteration, gradients are calculated to refine the new clusters from the previously created clusters [14]. The iterative process continues till the algorithm converges after the gradients stop changing, thus named gradient-ascent methods. A lot of research has been already done in the domain of gradient-ascent methods; the list of these methods is presented in Table 1.



**Table 1.** *Cont.*


**Table 1.** *Cont.*

The gradient ascent methods for super-pixel creation seems to be a promising solution due to their simplicity of implementation, speed of processing and easy adaptation for handling the latest demands of complex visual image scenarios. However, the concerns associated with underlying proper extraction strategies is one of the challenging aspects to cater the dynamic featural requirements imposed by complex visual image scenarios such as semi-dark images.

#### *2.2. Critical Analysis of Gradient-Ascent Super-Pixel Creation Algorithms Based on Manipulation Strategy*

For the critical analysis we have considered gradient ascent based super-pixel algorithms presented over period of two decades ranging from 2001 through 2021. The studies are retrieved from Google Scholar's repository with keywords including superpixel segmentation, pixel abstraction, content sensitive super-pixel creation, content-aware super-pixel segmentation. The search resulted in a lot of segmentation related studies in domain of image processing including the basic image transformations along with related super-pixel segmentation studies. For the critical analysis, the studies mentioning clustering based super-pixel creations were shortlisted due to their relevance with proposed algorithm. The key features of these studies are critically analyzed and comprehensively presented in Table 1, along with the critiques for respective handling concerns associated with semi-dark imagery.
