Article Preview
Top1. Introduction
Non-photorealistic rendering (NPR) is a promising technology since 19th century which emerged as a branch of image processing and computer graphics (Jan Eric Kyprianidis et al., 2013). Image abstraction and artistic stylization is considered as an advanced technology under the NPR domain and has been an effective visual tool for many applications. Image abstraction is a technique that reduces the superfluous details and preserves the significant information by protecting the salience gradient structure features of an image for storage. The advancement of science and technology, in fact, has broadened the scope of image abstraction. This process is a contrivance to articulate the nature of image content in an effective manner. Further, removal of irrelevant information in an image scene reduces the size (Image feature dimensionality) of an image thereby enhancing its clarity. Whereas, the stylization process involves creation of stylistic output from 2D abstracted color photographs. Image stylization is a process of creating the visually pleasing arts (sculpture, cinematography and architecture painting).Using non-photorealistic rendering filters, abstracted images are further refined into coherent artistic stylization images (Nagendra Swamy H.S and Kumar M.P.P, 2013).
In order to derive the best abstraction and artistic stylization impression rendering from low-illuminated (Kumar MPP et al., 2020) and underexposed images, a dire need is felt to adopt the filtering and high level image analysis techniques such as non-photorealistic rendering (NPR) image filtering, computer vision and image enhancement techniques. However, they demand a high computation facility to furnish the affluent features to abstraction and stylization. Hence, image abstraction and artistic stylization indeed has revolutionized the field of non-photorealistic rendering (NPR).
Improper lightening, lack of light reflections, inadequate lens and aperture settings during image capturing process lead to creation of underexposed images. In underexposed images the whole image color, contrast, edge strength and sharpness are too low which fails to convey the precise visual information. However, in most of the situations low-illumination and underexposed image enhancement depends on the fusion processes which consist of multiple images (P. Kumar and N. Swamy, 2013). In the fusion process, capturing the multiple images of same scene at different time and fused the series of infrared/ low-illuminated images to make the images more presentable such a kind of images are called as High Definition(HD) / High Dynamic Range images (HDR). Fusion processes involve either bright channel or dark channel correction thus it may not ensure the high quality output. When a high definition (HD)/high dynamic range (HDR) image is underexposed and low-illuminated, the surrounding region pixel values are almost near to 0 and when it is overexposed, the surrounding region pixel intensity values become negative.
In many contexts, images with underexposed complex backgrounds consist of diversified information and hence assessing it manually is difficult particularly for people with learning disabilities and low vision/visualization susceptance. The complexities of underexposed images are expressed in terms of patterns, shapes, color contrast and sharpness. It is difficult to identify and segregate the structure information as being dominant and non-dominant features. When dealing with low illumination images the problem becomes more challenging. Low-illumination images are usually captured during night time or poor lighting conditions leading to poor visibility. Information extraction in such cases is very much challenging and structural content recovery under this circumstance is most useful for the outdoor surveillance system (Shi, Z., et al., 2018) and computer vision applications. Hence, non-linearity of the blatant underexposed and low-illuminated images could be corrected and recovery of the image content is accomplished by adopting suitable tone mapping operators (TMO’s) along with NPR filtering techniques.