Evaluation of Image Quality Assessment Metrics: Color Quantization Noise

Abstract

Although color quantization noise is frequently met in practice, it has not been given too much attention in color image visual quality assessment. In this paper, a new image database for the evaluation of image quality metrics over color quantization noise is described. It contains 25 reference images and 875 test images produced by five popular quantization algorithms. Each of the quantized images was evaluated by 22 human subjects and more than 19200 individual human quality judgments were carried out to obtain the final mean opinion scores. A comparative analysis of several well-known image quality metrics is presented and their correlation with the human opinion scores is evaluated. This image database has been made freely available for downloading for research in image quality assessment and other applications [10]. General Terms:Subjective image quality assessment, Objective image quality assessment

Evaluation of quality measures for color quantization

“…CQD has been fully extracted and its data has been re-arranged so that it can be compared with TID. Since the considered metrics values were used for CQD in [27] but are not publicly available, these values have been recalculated using, as done in [26], the Matlab code Metrix MUX Visual Quality Assessment package [87] so to favorite the comparison with TID data. From TID, data have been extracted to form two sub-databases by selecting those relative to the color quantization distortion (#7) and color quantization followed by dithering (#22): -TID*: containing TID images with CQ distortion (#7) and the corresponding MOS; -TIDD*: containing TID images with CQ distortion plus dithering (#22) and the corresponding MOS.…”

“…Usually, the existing currently available databases include more than a distortion on which often the quality assessment is generally carried out globally by considering as equivalent different types of noise. Some exceptions are, for instance, the database proposed in [83] for blurring distortion and the aforementioned CQD database [27]. The specific subjective testing methodologies vary, but each image in the databases is labeled with a MOS, which represents the average subjective opinion about the quality of the image and is often referred to as the "ground truth" quality score of the image.…”

“…In this section, a brief description of these two databases, functional to the discussion of the evaluation process described in sections 4 and 5, is given. For more details regarding these databases see [26] and [27], respectively.…”

“…In particular, the distortion introduced by the CQ process has been considered initially in Tampere Image Database, namely TID2008 [25]. Successively the CQ distortions have been considered in TID2013 database [26], which is the largest so far available image database, and in Color Quantization Database, here indicated as CQD [27] (see section 2.5, 3.1, 3.2). Moreover, in IQA databases, quality measures based on color/HVS properties are not publicly available included in those where CQ distortion is considered.…”

[Paper] Noise Bias Compensation for Color Images after Tone Mapping

“…It is also used to accelerate color image processing tasks such as edge detection, image enhancement, histogram calculation, and color adjustment. However, there have been few studies on the reduction of color quantization noise [12]. Note that the mean of color quantization noise is not necessarily zero, and an image that has undergone color quantization and TM has NB.…”