First, we extracted general impressions about crinkle finishes. Fourteen subjects observed three crinkle-finished samples and two general DSLRs and then wrote down impressions they felt with single words or short phrases as many as they could. They were encouraged to clamp samples and view them at various angles. They could write down only visual impressions, not haptic impressions. The subjects were encouraged to observe a small flat area (about 15 mm × 20 mm) to avoid the effects of sample and camera shapes. The subjects were image science engineers in our company and consisted of 13 men and 1 woman in their 20s to 50s. They were selected randomly without any consideration of their knowledge of DSLR. They were told that samples were parts of DSLR exteriors.

This test was conducted using a standard light source (Spectra Light QC; X-Rite, MA, USA). Standard illuminant D65 was used, and the illuminance of the observation area was set at 1,300 lx. These conditions were also used for the appropriateness tests and impression evaluation experiments.

After all participants wrote down impressions, we took all words that had appeared at least once. Then, 114 words were extracted from this experiment. Additionally, we added 25 words used by paint designers. A total of 139 words were extracted to describe the general impressions of crinkle finish.

The words gathered by the impression-word extraction test contain complex and uncommon words. These words are not appropriate for use in evaluation experiments. Therefore, in this test, we evaluated appropriateness of these words for crinkle paint impression evaluation.

In this test, 21 participants evaluated the appropriateness of 139 words using seven scores from 1 (“very inappropriate”) to 7 (“very appropriate”). Subjects observed three crinkle-finished samples and two general DSLRs and then evaluated how appropriate each word was for the impression of samples. In this test, subjects could also clamp samples and observe them at various angles. Subjects were encouraged to observe a small flat area to avoid the effects of sample and camera shapes. Ten participants were instructed to evaluate how appropriate each word was for the impression of physical condition and 11 participants were instructed to evaluate how appropriate each word was for the impression of what they feel from samples. Because the number of evaluation words must be smaller than the number of samples to perform factor analyses, the number of evaluated impressions is limited. Therefore, we analyzed physical impressions and emotional impressions separately to evaluate more impressions [5, 6, 11, 21]. We call the former “low-level impressions” and the latter “high-level impressions.” We calculated the average and standard deviation of each sample score. Words with an average score of more than five and a standard deviation of less than 1.5 were selected as appropriate words.

We extracted 36 words appropriate for low-level impressions and 32 words appropriate for high-level impressions. Eighteen words were appropriate for both levels of impression.

As mentioned in Section 1, the crinkle finish is intended to provide a premium feeling. However, no words related to ‘feels premium’ were extracted by the impression-word appropriateness test. Hence, we added “feels premium” to the high-level impression appropriate words to investigate the relationship between the premium feeling and other impressions.

We conducted a rating word distance measurement test to measure the distances of the meanings between impression words.

Participants were handed the list of all pairs of words extracted in the previous experiment and judged whether each pair of words could be exchanged semantically. This experiment was conducted without the observation of samples. Ten participants evaluated pairs of low-level impression words. They had also evaluated appropriateness of low-level impression at the impression-word appropriateness test. Twelve participants evaluated pairs of high-level impression words. Among 12 participants, 11 participants had also evaluated appropriateness of high-level impression at the impression-word appropriateness test and 1 was a new participant. The new participant was shown crinkle-finished samples before this experiment.

Then, the word distances between each pair were calculated by the following equation: (1)D=NnNall. Here, D is the word distance between a pair of words, N_all is the number of all participants, and N_n is the number of participants who judged that the pair of words could not be exchanged semantically. When all participants answer that a pair of words cannot be exchanged, D = 1. When all participants answer that a pair of words can be exchanged, D = 0. Therefore, D corresponds to normalized similarity between a pair of appropriate words. A pair with a short distance indicates a similar impression and a pair with a long distance indicates different impressions.

We plotted the words in multi-dimensional space using multi-dimensional scaling. The maximum number of dimensions of the space is one less than the number of appropriate words. To analyze the similarities between appropriate words correctly, low- and high-level impression words were plotted in 35- and 31-dimensional spaces, respectively.

To group impression words according to the similarity of impressions, hierarchical cluster analysis was conducted using Ward’s method. The high- and low-level impression words were divided into nine clusters, each corresponding to one type of impression. The nine impression words that were the nearest to the center of mass of each cluster were selected as the final evaluation words (Table I).

Figure 2 shows an example of the crinkle paint samples used in the impression evaluation experiments. The samples are parts of camera exteriors that were finished using crinkle paints. Table II shows the quantity and size of the paint particles used for the 10 crinkle-finished samples. Three levels of each factor were tested. The sizes of large particles are approximately 0.08 mm², medium particles are 0.06 mm², and small particles are 0.05 mm². Furthermore, we used two particle coating materials (TYPE-A and TYPE-B). The base coating material is the same for all samples. Figure 3 shows some examples of pictures of sample surfaces.

Participants were presented each sample separately and evaluated how strong they felt each impression, which is extracted by a rating word distance measurement test from crinkle-finished samples on a five-level scale from “not felt at all” to “strongly felt.” They were encouraged to clamp samples and view them at various angles. They were encouraged to evaluate only visual impressions, not haptic impressions. The samples were shown in random order. Eleven participants evaluated low-level impressions. Among 11 participants, 10 participants had also evaluated low-level impressions in previous experiments and 1 was a new participant. Ten participants evaluated high-level impressions. They had also evaluated high-level impressions in previous experiments.

Macroroughness is estimated as follows: (4)Macroroughness =0.978×x11. Here, x₁₁ is the 45∘∕45∘ high spatial frequency characteristic. Figure 10 shows the relationship between the values predicted from Eq. (4) and the subjective scores for macroroughness obtained by the impression evaluation experiments.

Equation (4) indicates that macroroughness is highly positively correlated with the 45∘∕45∘ high spatial frequency characteristic. The R-squared value of Eq. (4) is 0.95. By k-fold cross-validation (k = 11), a mean R-squared value of 0.924 was obtained.

Microroughness is estimated as follows: (5)Microroughness =0.529×x21−0.51×x22−1.03×x23. Here, x₂₁ is the 45∘∕0∘ standard deviation of L^∗, x₂₂ is the 60∘∕45∘ base value of L^∗, and x₂₃ is the 0∘∕45∘ average particle area. Figure 11 shows the relationship between the values predicted from Eq. (5) and the subjective scores.

Equation (5) indicates that microroughness is positively correlated with the standard deviation of L^∗ and negatively correlated with the base value of L^∗ values and the average particle area. The R-squared value is 0.863. By k-fold cross-validation (k = 11), a mean R-squared value of 0.664 was obtained.

The following equation estimates randomness: (6)Randomness =0.976×x31.

Here, x₃₁ is the 45∘∕30∘ average particle area. Figure 12 shows the relationship between the values predicted from (Eq. (6)) and the subjective scores.

Equation (6) indicates that randomness is positively correlated with particle areas. The R-squared value is 0.961. By k-fold cross-validation (k = 10), a mean R-squared value of 0.937 was obtained.

Stateliness can be estimated as follows: (7)Stateliness =0.701×x41+0.771×x42−0.811×x43. Here, x₄₁ is the 45∘∕0∘ low spatial frequency characteristic, x₄₂ is the 30∘∕60∘ standard deviation of L^∗, and x₄₃ is the 60∘∕60∘ standard deviation of L^∗. Figure 13 shows the relationship between the values predicted from Eq. (7) and the subjective scores.

Equation (7) indicates that stateliness is positively correlated with low spatial frequency characteristics and the 30∘∕60∘ standard deviation of L^∗. Furthermore, it is negatively correlated with the 60∘∕60∘ standard deviation of L^∗. The R-squared value is 0.973. By k-fold cross-validation (k = 10), a mean R-squared value of 0.921 was obtained.

Fineness is estimated as follows: (8)Fineness =1.817×x51−1.582×x52−1.676×x53. Here, x₅₁ is the 0∘∕45∘ contrast of L^∗, x₅₂ is the 0∘∕45∘ standard deviation of L^∗, and x₅₃ is the 0∘∕60∘ contrast of L^∗. Figure 14 shows the relationship between the values predicted from Eq. (8) and the subjective scores.

Equation (8) indicates that fineness is positively correlated with the 0∘∕45∘ contrast of L^∗. Furthermore, it is negatively correlated with the 0∘∕45∘ standard deviation of L^∗ and the 0∘∕60∘ contrast of L^∗. The R-squared value is 0.957. By k-fold cross-validation (k = 10), a mean R-squared value of 0.937 was obtained.